xref: /linux/net/core/sock.c (revision d69eb204c255c35abd9e8cb621484e8074c75eaa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Generic socket support routines. Memory allocators, socket lock/release
8  *		handler for protocols to use and generic option handler.
9  *
10  * Authors:	Ross Biro
11  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *		Florian La Roche, <flla@stud.uni-sb.de>
13  *		Alan Cox, <A.Cox@swansea.ac.uk>
14  *
15  * Fixes:
16  *		Alan Cox	: 	Numerous verify_area() problems
17  *		Alan Cox	:	Connecting on a connecting socket
18  *					now returns an error for tcp.
19  *		Alan Cox	:	sock->protocol is set correctly.
20  *					and is not sometimes left as 0.
21  *		Alan Cox	:	connect handles icmp errors on a
22  *					connect properly. Unfortunately there
23  *					is a restart syscall nasty there. I
24  *					can't match BSD without hacking the C
25  *					library. Ideas urgently sought!
26  *		Alan Cox	:	Disallow bind() to addresses that are
27  *					not ours - especially broadcast ones!!
28  *		Alan Cox	:	Socket 1024 _IS_ ok for users. (fencepost)
29  *		Alan Cox	:	sock_wfree/sock_rfree don't destroy sockets,
30  *					instead they leave that for the DESTROY timer.
31  *		Alan Cox	:	Clean up error flag in accept
32  *		Alan Cox	:	TCP ack handling is buggy, the DESTROY timer
33  *					was buggy. Put a remove_sock() in the handler
34  *					for memory when we hit 0. Also altered the timer
35  *					code. The ACK stuff can wait and needs major
36  *					TCP layer surgery.
37  *		Alan Cox	:	Fixed TCP ack bug, removed remove sock
38  *					and fixed timer/inet_bh race.
39  *		Alan Cox	:	Added zapped flag for TCP
40  *		Alan Cox	:	Move kfree_skb into skbuff.c and tidied up surplus code
41  *		Alan Cox	:	for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42  *		Alan Cox	:	kfree_s calls now are kfree_skbmem so we can track skb resources
43  *		Alan Cox	:	Supports socket option broadcast now as does udp. Packet and raw need fixing.
44  *		Alan Cox	:	Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45  *		Rick Sladkey	:	Relaxed UDP rules for matching packets.
46  *		C.E.Hawkins	:	IFF_PROMISC/SIOCGHWADDR support
47  *	Pauline Middelink	:	identd support
48  *		Alan Cox	:	Fixed connect() taking signals I think.
49  *		Alan Cox	:	SO_LINGER supported
50  *		Alan Cox	:	Error reporting fixes
51  *		Anonymous	:	inet_create tidied up (sk->reuse setting)
52  *		Alan Cox	:	inet sockets don't set sk->type!
53  *		Alan Cox	:	Split socket option code
54  *		Alan Cox	:	Callbacks
55  *		Alan Cox	:	Nagle flag for Charles & Johannes stuff
56  *		Alex		:	Removed restriction on inet fioctl
57  *		Alan Cox	:	Splitting INET from NET core
58  *		Alan Cox	:	Fixed bogus SO_TYPE handling in getsockopt()
59  *		Adam Caldwell	:	Missing return in SO_DONTROUTE/SO_DEBUG code
60  *		Alan Cox	:	Split IP from generic code
61  *		Alan Cox	:	New kfree_skbmem()
62  *		Alan Cox	:	Make SO_DEBUG superuser only.
63  *		Alan Cox	:	Allow anyone to clear SO_DEBUG
64  *					(compatibility fix)
65  *		Alan Cox	:	Added optimistic memory grabbing for AF_UNIX throughput.
66  *		Alan Cox	:	Allocator for a socket is settable.
67  *		Alan Cox	:	SO_ERROR includes soft errors.
68  *		Alan Cox	:	Allow NULL arguments on some SO_ opts
69  *		Alan Cox	: 	Generic socket allocation to make hooks
70  *					easier (suggested by Craig Metz).
71  *		Michael Pall	:	SO_ERROR returns positive errno again
72  *              Steve Whitehouse:       Added default destructor to free
73  *                                      protocol private data.
74  *              Steve Whitehouse:       Added various other default routines
75  *                                      common to several socket families.
76  *              Chris Evans     :       Call suser() check last on F_SETOWN
77  *		Jay Schulist	:	Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78  *		Andi Kleen	:	Add sock_kmalloc()/sock_kfree_s()
79  *		Andi Kleen	:	Fix write_space callback
80  *		Chris Evans	:	Security fixes - signedness again
81  *		Arnaldo C. Melo :       cleanups, use skb_queue_purge
82  *
83  * To Fix:
84  */
85 
86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
87 
88 #include <linux/unaligned.h>
89 #include <linux/capability.h>
90 #include <linux/errno.h>
91 #include <linux/errqueue.h>
92 #include <linux/types.h>
93 #include <linux/socket.h>
94 #include <linux/in.h>
95 #include <linux/kernel.h>
96 #include <linux/module.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/sched.h>
100 #include <linux/sched/mm.h>
101 #include <linux/timer.h>
102 #include <linux/string.h>
103 #include <linux/sockios.h>
104 #include <linux/net.h>
105 #include <linux/mm.h>
106 #include <linux/slab.h>
107 #include <linux/interrupt.h>
108 #include <linux/poll.h>
109 #include <linux/tcp.h>
110 #include <linux/udp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.h>
114 #include <linux/static_key.h>
115 #include <linux/memcontrol.h>
116 #include <linux/prefetch.h>
117 #include <linux/compat.h>
118 #include <linux/mroute.h>
119 #include <linux/mroute6.h>
120 #include <linux/icmpv6.h>
121 
122 #include <linux/uaccess.h>
123 
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <linux/skbuff_ref.h>
128 #include <net/net_namespace.h>
129 #include <net/request_sock.h>
130 #include <net/sock.h>
131 #include <net/proto_memory.h>
132 #include <linux/net_tstamp.h>
133 #include <net/xfrm.h>
134 #include <linux/ipsec.h>
135 #include <net/cls_cgroup.h>
136 #include <net/netprio_cgroup.h>
137 #include <linux/sock_diag.h>
138 
139 #include <linux/filter.h>
140 #include <net/sock_reuseport.h>
141 #include <net/bpf_sk_storage.h>
142 
143 #include <trace/events/sock.h>
144 
145 #include <net/tcp.h>
146 #include <net/busy_poll.h>
147 #include <net/phonet/phonet.h>
148 
149 #include <linux/ethtool.h>
150 
151 #include <uapi/linux/pidfd.h>
152 
153 #include "dev.h"
154 
155 static DEFINE_MUTEX(proto_list_mutex);
156 static LIST_HEAD(proto_list);
157 
158 static void sock_def_write_space_wfree(struct sock *sk);
159 static void sock_def_write_space(struct sock *sk);
160 
161 /**
162  * sk_ns_capable - General socket capability test
163  * @sk: Socket to use a capability on or through
164  * @user_ns: The user namespace of the capability to use
165  * @cap: The capability to use
166  *
167  * Test to see if the opener of the socket had when the socket was
168  * created and the current process has the capability @cap in the user
169  * namespace @user_ns.
170  */
171 bool sk_ns_capable(const struct sock *sk,
172 		   struct user_namespace *user_ns, int cap)
173 {
174 	return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
175 		ns_capable(user_ns, cap);
176 }
177 EXPORT_SYMBOL(sk_ns_capable);
178 
179 /**
180  * sk_capable - Socket global capability test
181  * @sk: Socket to use a capability on or through
182  * @cap: The global capability to use
183  *
184  * Test to see if the opener of the socket had when the socket was
185  * created and the current process has the capability @cap in all user
186  * namespaces.
187  */
188 bool sk_capable(const struct sock *sk, int cap)
189 {
190 	return sk_ns_capable(sk, &init_user_ns, cap);
191 }
192 EXPORT_SYMBOL(sk_capable);
193 
194 /**
195  * sk_net_capable - Network namespace socket capability test
196  * @sk: Socket to use a capability on or through
197  * @cap: The capability to use
198  *
199  * Test to see if the opener of the socket had when the socket was created
200  * and the current process has the capability @cap over the network namespace
201  * the socket is a member of.
202  */
203 bool sk_net_capable(const struct sock *sk, int cap)
204 {
205 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
206 }
207 EXPORT_SYMBOL(sk_net_capable);
208 
209 /*
210  * Each address family might have different locking rules, so we have
211  * one slock key per address family and separate keys for internal and
212  * userspace sockets.
213  */
214 static struct lock_class_key af_family_keys[AF_MAX];
215 static struct lock_class_key af_family_kern_keys[AF_MAX];
216 static struct lock_class_key af_family_slock_keys[AF_MAX];
217 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
218 
219 /*
220  * Make lock validator output more readable. (we pre-construct these
221  * strings build-time, so that runtime initialization of socket
222  * locks is fast):
223  */
224 
225 #define _sock_locks(x)						  \
226   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
227   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
228   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
229   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
230   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
231   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
232   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
233   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
234   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
235   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
236   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
237   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
238   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
239   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
240   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_XDP"	, \
241   x "AF_MCTP"  , \
242   x "AF_MAX"
243 
244 static const char *const af_family_key_strings[AF_MAX+1] = {
245 	_sock_locks("sk_lock-")
246 };
247 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
248 	_sock_locks("slock-")
249 };
250 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
251 	_sock_locks("clock-")
252 };
253 
254 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
255 	_sock_locks("k-sk_lock-")
256 };
257 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
258 	_sock_locks("k-slock-")
259 };
260 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
261 	_sock_locks("k-clock-")
262 };
263 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
264 	_sock_locks("rlock-")
265 };
266 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
267 	_sock_locks("wlock-")
268 };
269 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
270 	_sock_locks("elock-")
271 };
272 
273 /*
274  * sk_callback_lock and sk queues locking rules are per-address-family,
275  * so split the lock classes by using a per-AF key:
276  */
277 static struct lock_class_key af_callback_keys[AF_MAX];
278 static struct lock_class_key af_rlock_keys[AF_MAX];
279 static struct lock_class_key af_wlock_keys[AF_MAX];
280 static struct lock_class_key af_elock_keys[AF_MAX];
281 static struct lock_class_key af_kern_callback_keys[AF_MAX];
282 
283 /* Run time adjustable parameters. */
284 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
285 EXPORT_SYMBOL(sysctl_wmem_max);
286 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
287 EXPORT_SYMBOL(sysctl_rmem_max);
288 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
289 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
290 
291 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key);
292 EXPORT_SYMBOL_GPL(memalloc_socks_key);
293 
294 /**
295  * sk_set_memalloc - sets %SOCK_MEMALLOC
296  * @sk: socket to set it on
297  *
298  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
299  * It's the responsibility of the admin to adjust min_free_kbytes
300  * to meet the requirements
301  */
302 void sk_set_memalloc(struct sock *sk)
303 {
304 	sock_set_flag(sk, SOCK_MEMALLOC);
305 	sk->sk_allocation |= __GFP_MEMALLOC;
306 	static_branch_inc(&memalloc_socks_key);
307 }
308 EXPORT_SYMBOL_GPL(sk_set_memalloc);
309 
310 void sk_clear_memalloc(struct sock *sk)
311 {
312 	sock_reset_flag(sk, SOCK_MEMALLOC);
313 	sk->sk_allocation &= ~__GFP_MEMALLOC;
314 	static_branch_dec(&memalloc_socks_key);
315 
316 	/*
317 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
318 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
319 	 * it has rmem allocations due to the last swapfile being deactivated
320 	 * but there is a risk that the socket is unusable due to exceeding
321 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
322 	 */
323 	sk_mem_reclaim(sk);
324 }
325 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
326 
327 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
328 {
329 	int ret;
330 	unsigned int noreclaim_flag;
331 
332 	/* these should have been dropped before queueing */
333 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
334 
335 	noreclaim_flag = memalloc_noreclaim_save();
336 	ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv,
337 				 tcp_v6_do_rcv,
338 				 tcp_v4_do_rcv,
339 				 sk, skb);
340 	memalloc_noreclaim_restore(noreclaim_flag);
341 
342 	return ret;
343 }
344 EXPORT_SYMBOL(__sk_backlog_rcv);
345 
346 void sk_error_report(struct sock *sk)
347 {
348 	sk->sk_error_report(sk);
349 
350 	switch (sk->sk_family) {
351 	case AF_INET:
352 		fallthrough;
353 	case AF_INET6:
354 		trace_inet_sk_error_report(sk);
355 		break;
356 	default:
357 		break;
358 	}
359 }
360 EXPORT_SYMBOL(sk_error_report);
361 
362 int sock_get_timeout(long timeo, void *optval, bool old_timeval)
363 {
364 	struct __kernel_sock_timeval tv;
365 
366 	if (timeo == MAX_SCHEDULE_TIMEOUT) {
367 		tv.tv_sec = 0;
368 		tv.tv_usec = 0;
369 	} else {
370 		tv.tv_sec = timeo / HZ;
371 		tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ;
372 	}
373 
374 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
375 		struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec };
376 		*(struct old_timeval32 *)optval = tv32;
377 		return sizeof(tv32);
378 	}
379 
380 	if (old_timeval) {
381 		struct __kernel_old_timeval old_tv;
382 		old_tv.tv_sec = tv.tv_sec;
383 		old_tv.tv_usec = tv.tv_usec;
384 		*(struct __kernel_old_timeval *)optval = old_tv;
385 		return sizeof(old_tv);
386 	}
387 
388 	*(struct __kernel_sock_timeval *)optval = tv;
389 	return sizeof(tv);
390 }
391 EXPORT_SYMBOL(sock_get_timeout);
392 
393 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
394 			   sockptr_t optval, int optlen, bool old_timeval)
395 {
396 	if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) {
397 		struct old_timeval32 tv32;
398 
399 		if (optlen < sizeof(tv32))
400 			return -EINVAL;
401 
402 		if (copy_from_sockptr(&tv32, optval, sizeof(tv32)))
403 			return -EFAULT;
404 		tv->tv_sec = tv32.tv_sec;
405 		tv->tv_usec = tv32.tv_usec;
406 	} else if (old_timeval) {
407 		struct __kernel_old_timeval old_tv;
408 
409 		if (optlen < sizeof(old_tv))
410 			return -EINVAL;
411 		if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv)))
412 			return -EFAULT;
413 		tv->tv_sec = old_tv.tv_sec;
414 		tv->tv_usec = old_tv.tv_usec;
415 	} else {
416 		if (optlen < sizeof(*tv))
417 			return -EINVAL;
418 		if (copy_from_sockptr(tv, optval, sizeof(*tv)))
419 			return -EFAULT;
420 	}
421 
422 	return 0;
423 }
424 EXPORT_SYMBOL(sock_copy_user_timeval);
425 
426 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen,
427 			    bool old_timeval)
428 {
429 	struct __kernel_sock_timeval tv;
430 	int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval);
431 	long val;
432 
433 	if (err)
434 		return err;
435 
436 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
437 		return -EDOM;
438 
439 	if (tv.tv_sec < 0) {
440 		static int warned __read_mostly;
441 
442 		WRITE_ONCE(*timeo_p, 0);
443 		if (warned < 10 && net_ratelimit()) {
444 			warned++;
445 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
446 				__func__, current->comm, task_pid_nr(current));
447 		}
448 		return 0;
449 	}
450 	val = MAX_SCHEDULE_TIMEOUT;
451 	if ((tv.tv_sec || tv.tv_usec) &&
452 	    (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)))
453 		val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec,
454 						    USEC_PER_SEC / HZ);
455 	WRITE_ONCE(*timeo_p, val);
456 	return 0;
457 }
458 
459 static bool sk_set_prio_allowed(const struct sock *sk, int val)
460 {
461 	return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) ||
462 		sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) ||
463 		sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN));
464 }
465 
466 static bool sock_needs_netstamp(const struct sock *sk)
467 {
468 	switch (sk->sk_family) {
469 	case AF_UNSPEC:
470 	case AF_UNIX:
471 		return false;
472 	default:
473 		return true;
474 	}
475 }
476 
477 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
478 {
479 	if (sk->sk_flags & flags) {
480 		sk->sk_flags &= ~flags;
481 		if (sock_needs_netstamp(sk) &&
482 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
483 			net_disable_timestamp();
484 	}
485 }
486 
487 
488 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
489 {
490 	unsigned long flags;
491 	struct sk_buff_head *list = &sk->sk_receive_queue;
492 
493 	if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) {
494 		atomic_inc(&sk->sk_drops);
495 		trace_sock_rcvqueue_full(sk, skb);
496 		return -ENOMEM;
497 	}
498 
499 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
500 		atomic_inc(&sk->sk_drops);
501 		return -ENOBUFS;
502 	}
503 
504 	skb->dev = NULL;
505 	skb_set_owner_r(skb, sk);
506 
507 	/* we escape from rcu protected region, make sure we dont leak
508 	 * a norefcounted dst
509 	 */
510 	skb_dst_force(skb);
511 
512 	spin_lock_irqsave(&list->lock, flags);
513 	sock_skb_set_dropcount(sk, skb);
514 	__skb_queue_tail(list, skb);
515 	spin_unlock_irqrestore(&list->lock, flags);
516 
517 	if (!sock_flag(sk, SOCK_DEAD))
518 		sk->sk_data_ready(sk);
519 	return 0;
520 }
521 EXPORT_SYMBOL(__sock_queue_rcv_skb);
522 
523 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
524 			      enum skb_drop_reason *reason)
525 {
526 	enum skb_drop_reason drop_reason;
527 	int err;
528 
529 	err = sk_filter_reason(sk, skb, &drop_reason);
530 	if (err)
531 		goto out;
532 
533 	err = __sock_queue_rcv_skb(sk, skb);
534 	switch (err) {
535 	case -ENOMEM:
536 		drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
537 		break;
538 	case -ENOBUFS:
539 		drop_reason = SKB_DROP_REASON_PROTO_MEM;
540 		break;
541 	default:
542 		drop_reason = SKB_NOT_DROPPED_YET;
543 		break;
544 	}
545 out:
546 	if (reason)
547 		*reason = drop_reason;
548 	return err;
549 }
550 EXPORT_SYMBOL(sock_queue_rcv_skb_reason);
551 
552 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
553 		     const int nested, unsigned int trim_cap, bool refcounted)
554 {
555 	enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
556 	int rc = NET_RX_SUCCESS;
557 	int err;
558 
559 	if (sk_filter_trim_cap(sk, skb, trim_cap, &reason))
560 		goto discard_and_relse;
561 
562 	skb->dev = NULL;
563 
564 	if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) {
565 		atomic_inc(&sk->sk_drops);
566 		reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
567 		goto discard_and_relse;
568 	}
569 	if (nested)
570 		bh_lock_sock_nested(sk);
571 	else
572 		bh_lock_sock(sk);
573 	if (!sock_owned_by_user(sk)) {
574 		/*
575 		 * trylock + unlock semantics:
576 		 */
577 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
578 
579 		rc = sk_backlog_rcv(sk, skb);
580 
581 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
582 	} else if ((err = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))) {
583 		bh_unlock_sock(sk);
584 		if (err == -ENOMEM)
585 			reason = SKB_DROP_REASON_PFMEMALLOC;
586 		if (err == -ENOBUFS)
587 			reason = SKB_DROP_REASON_SOCKET_BACKLOG;
588 		atomic_inc(&sk->sk_drops);
589 		goto discard_and_relse;
590 	}
591 
592 	bh_unlock_sock(sk);
593 out:
594 	if (refcounted)
595 		sock_put(sk);
596 	return rc;
597 discard_and_relse:
598 	sk_skb_reason_drop(sk, skb, reason);
599 	goto out;
600 }
601 EXPORT_SYMBOL(__sk_receive_skb);
602 
603 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *,
604 							  u32));
605 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
606 							   u32));
607 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
608 {
609 	struct dst_entry *dst = __sk_dst_get(sk);
610 
611 	if (dst && READ_ONCE(dst->obsolete) &&
612 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
613 			       dst, cookie) == NULL) {
614 		sk_tx_queue_clear(sk);
615 		WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
616 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
617 		dst_release(dst);
618 		return NULL;
619 	}
620 
621 	return dst;
622 }
623 EXPORT_SYMBOL(__sk_dst_check);
624 
625 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
626 {
627 	struct dst_entry *dst = sk_dst_get(sk);
628 
629 	if (dst && READ_ONCE(dst->obsolete) &&
630 	    INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check,
631 			       dst, cookie) == NULL) {
632 		sk_dst_reset(sk);
633 		dst_release(dst);
634 		return NULL;
635 	}
636 
637 	return dst;
638 }
639 EXPORT_SYMBOL(sk_dst_check);
640 
641 static int sock_bindtoindex_locked(struct sock *sk, int ifindex)
642 {
643 	int ret = -ENOPROTOOPT;
644 #ifdef CONFIG_NETDEVICES
645 	struct net *net = sock_net(sk);
646 
647 	/* Sorry... */
648 	ret = -EPERM;
649 	if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW))
650 		goto out;
651 
652 	ret = -EINVAL;
653 	if (ifindex < 0)
654 		goto out;
655 
656 	/* Paired with all READ_ONCE() done locklessly. */
657 	WRITE_ONCE(sk->sk_bound_dev_if, ifindex);
658 
659 	if (sk->sk_prot->rehash)
660 		sk->sk_prot->rehash(sk);
661 	sk_dst_reset(sk);
662 
663 	ret = 0;
664 
665 out:
666 #endif
667 
668 	return ret;
669 }
670 
671 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk)
672 {
673 	int ret;
674 
675 	if (lock_sk)
676 		lock_sock(sk);
677 	ret = sock_bindtoindex_locked(sk, ifindex);
678 	if (lock_sk)
679 		release_sock(sk);
680 
681 	return ret;
682 }
683 EXPORT_SYMBOL(sock_bindtoindex);
684 
685 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen)
686 {
687 	int ret = -ENOPROTOOPT;
688 #ifdef CONFIG_NETDEVICES
689 	struct net *net = sock_net(sk);
690 	char devname[IFNAMSIZ];
691 	int index;
692 
693 	ret = -EINVAL;
694 	if (optlen < 0)
695 		goto out;
696 
697 	/* Bind this socket to a particular device like "eth0",
698 	 * as specified in the passed interface name. If the
699 	 * name is "" or the option length is zero the socket
700 	 * is not bound.
701 	 */
702 	if (optlen > IFNAMSIZ - 1)
703 		optlen = IFNAMSIZ - 1;
704 	memset(devname, 0, sizeof(devname));
705 
706 	ret = -EFAULT;
707 	if (copy_from_sockptr(devname, optval, optlen))
708 		goto out;
709 
710 	index = 0;
711 	if (devname[0] != '\0') {
712 		struct net_device *dev;
713 
714 		rcu_read_lock();
715 		dev = dev_get_by_name_rcu(net, devname);
716 		if (dev)
717 			index = dev->ifindex;
718 		rcu_read_unlock();
719 		ret = -ENODEV;
720 		if (!dev)
721 			goto out;
722 	}
723 
724 	sockopt_lock_sock(sk);
725 	ret = sock_bindtoindex_locked(sk, index);
726 	sockopt_release_sock(sk);
727 out:
728 #endif
729 
730 	return ret;
731 }
732 
733 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval,
734 				sockptr_t optlen, int len)
735 {
736 	int ret = -ENOPROTOOPT;
737 #ifdef CONFIG_NETDEVICES
738 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
739 	struct net *net = sock_net(sk);
740 	char devname[IFNAMSIZ];
741 
742 	if (bound_dev_if == 0) {
743 		len = 0;
744 		goto zero;
745 	}
746 
747 	ret = -EINVAL;
748 	if (len < IFNAMSIZ)
749 		goto out;
750 
751 	ret = netdev_get_name(net, devname, bound_dev_if);
752 	if (ret)
753 		goto out;
754 
755 	len = strlen(devname) + 1;
756 
757 	ret = -EFAULT;
758 	if (copy_to_sockptr(optval, devname, len))
759 		goto out;
760 
761 zero:
762 	ret = -EFAULT;
763 	if (copy_to_sockptr(optlen, &len, sizeof(int)))
764 		goto out;
765 
766 	ret = 0;
767 
768 out:
769 #endif
770 
771 	return ret;
772 }
773 
774 bool sk_mc_loop(const struct sock *sk)
775 {
776 	if (dev_recursion_level())
777 		return false;
778 	if (!sk)
779 		return true;
780 	/* IPV6_ADDRFORM can change sk->sk_family under us. */
781 	switch (READ_ONCE(sk->sk_family)) {
782 	case AF_INET:
783 		return inet_test_bit(MC_LOOP, sk);
784 #if IS_ENABLED(CONFIG_IPV6)
785 	case AF_INET6:
786 		return inet6_test_bit(MC6_LOOP, sk);
787 #endif
788 	}
789 	WARN_ON_ONCE(1);
790 	return true;
791 }
792 EXPORT_SYMBOL(sk_mc_loop);
793 
794 void sock_set_reuseaddr(struct sock *sk)
795 {
796 	lock_sock(sk);
797 	sk->sk_reuse = SK_CAN_REUSE;
798 	release_sock(sk);
799 }
800 EXPORT_SYMBOL(sock_set_reuseaddr);
801 
802 void sock_set_reuseport(struct sock *sk)
803 {
804 	lock_sock(sk);
805 	sk->sk_reuseport = true;
806 	release_sock(sk);
807 }
808 EXPORT_SYMBOL(sock_set_reuseport);
809 
810 void sock_no_linger(struct sock *sk)
811 {
812 	lock_sock(sk);
813 	WRITE_ONCE(sk->sk_lingertime, 0);
814 	sock_set_flag(sk, SOCK_LINGER);
815 	release_sock(sk);
816 }
817 EXPORT_SYMBOL(sock_no_linger);
818 
819 void sock_set_priority(struct sock *sk, u32 priority)
820 {
821 	WRITE_ONCE(sk->sk_priority, priority);
822 }
823 EXPORT_SYMBOL(sock_set_priority);
824 
825 void sock_set_sndtimeo(struct sock *sk, s64 secs)
826 {
827 	if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1)
828 		WRITE_ONCE(sk->sk_sndtimeo, secs * HZ);
829 	else
830 		WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT);
831 }
832 EXPORT_SYMBOL(sock_set_sndtimeo);
833 
834 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns)
835 {
836 	sock_valbool_flag(sk, SOCK_RCVTSTAMP, val);
837 	sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns);
838 	if (val)  {
839 		sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new);
840 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
841 	}
842 }
843 
844 void sock_set_timestamp(struct sock *sk, int optname, bool valbool)
845 {
846 	switch (optname) {
847 	case SO_TIMESTAMP_OLD:
848 		__sock_set_timestamps(sk, valbool, false, false);
849 		break;
850 	case SO_TIMESTAMP_NEW:
851 		__sock_set_timestamps(sk, valbool, true, false);
852 		break;
853 	case SO_TIMESTAMPNS_OLD:
854 		__sock_set_timestamps(sk, valbool, false, true);
855 		break;
856 	case SO_TIMESTAMPNS_NEW:
857 		__sock_set_timestamps(sk, valbool, true, true);
858 		break;
859 	}
860 }
861 
862 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index)
863 {
864 	struct net *net = sock_net(sk);
865 	struct net_device *dev = NULL;
866 	bool match = false;
867 	int *vclock_index;
868 	int i, num;
869 
870 	if (sk->sk_bound_dev_if)
871 		dev = dev_get_by_index(net, sk->sk_bound_dev_if);
872 
873 	if (!dev) {
874 		pr_err("%s: sock not bind to device\n", __func__);
875 		return -EOPNOTSUPP;
876 	}
877 
878 	num = ethtool_get_phc_vclocks(dev, &vclock_index);
879 	dev_put(dev);
880 
881 	for (i = 0; i < num; i++) {
882 		if (*(vclock_index + i) == phc_index) {
883 			match = true;
884 			break;
885 		}
886 	}
887 
888 	if (num > 0)
889 		kfree(vclock_index);
890 
891 	if (!match)
892 		return -EINVAL;
893 
894 	WRITE_ONCE(sk->sk_bind_phc, phc_index);
895 
896 	return 0;
897 }
898 
899 int sock_set_timestamping(struct sock *sk, int optname,
900 			  struct so_timestamping timestamping)
901 {
902 	int val = timestamping.flags;
903 	int ret;
904 
905 	if (val & ~SOF_TIMESTAMPING_MASK)
906 		return -EINVAL;
907 
908 	if (val & SOF_TIMESTAMPING_OPT_ID_TCP &&
909 	    !(val & SOF_TIMESTAMPING_OPT_ID))
910 		return -EINVAL;
911 
912 	if (val & SOF_TIMESTAMPING_OPT_ID &&
913 	    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
914 		if (sk_is_tcp(sk)) {
915 			if ((1 << sk->sk_state) &
916 			    (TCPF_CLOSE | TCPF_LISTEN))
917 				return -EINVAL;
918 			if (val & SOF_TIMESTAMPING_OPT_ID_TCP)
919 				atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq);
920 			else
921 				atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una);
922 		} else {
923 			atomic_set(&sk->sk_tskey, 0);
924 		}
925 	}
926 
927 	if (val & SOF_TIMESTAMPING_OPT_STATS &&
928 	    !(val & SOF_TIMESTAMPING_OPT_TSONLY))
929 		return -EINVAL;
930 
931 	if (val & SOF_TIMESTAMPING_BIND_PHC) {
932 		ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc);
933 		if (ret)
934 			return ret;
935 	}
936 
937 	WRITE_ONCE(sk->sk_tsflags, val);
938 	sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW);
939 	sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY));
940 
941 	if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
942 		sock_enable_timestamp(sk,
943 				      SOCK_TIMESTAMPING_RX_SOFTWARE);
944 	else
945 		sock_disable_timestamp(sk,
946 				       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
947 	return 0;
948 }
949 
950 #if defined(CONFIG_CGROUP_BPF)
951 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
952 {
953 	struct bpf_sock_ops_kern sock_ops;
954 
955 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
956 	sock_ops.op = op;
957 	sock_ops.is_fullsock = 1;
958 	sock_ops.sk = sk;
959 	bpf_skops_init_skb(&sock_ops, skb, 0);
960 	__cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS);
961 }
962 #endif
963 
964 void sock_set_keepalive(struct sock *sk)
965 {
966 	lock_sock(sk);
967 	if (sk->sk_prot->keepalive)
968 		sk->sk_prot->keepalive(sk, true);
969 	sock_valbool_flag(sk, SOCK_KEEPOPEN, true);
970 	release_sock(sk);
971 }
972 EXPORT_SYMBOL(sock_set_keepalive);
973 
974 static void __sock_set_rcvbuf(struct sock *sk, int val)
975 {
976 	/* Ensure val * 2 fits into an int, to prevent max_t() from treating it
977 	 * as a negative value.
978 	 */
979 	val = min_t(int, val, INT_MAX / 2);
980 	sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
981 
982 	/* We double it on the way in to account for "struct sk_buff" etc.
983 	 * overhead.   Applications assume that the SO_RCVBUF setting they make
984 	 * will allow that much actual data to be received on that socket.
985 	 *
986 	 * Applications are unaware that "struct sk_buff" and other overheads
987 	 * allocate from the receive buffer during socket buffer allocation.
988 	 *
989 	 * And after considering the possible alternatives, returning the value
990 	 * we actually used in getsockopt is the most desirable behavior.
991 	 */
992 	WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF));
993 }
994 
995 void sock_set_rcvbuf(struct sock *sk, int val)
996 {
997 	lock_sock(sk);
998 	__sock_set_rcvbuf(sk, val);
999 	release_sock(sk);
1000 }
1001 EXPORT_SYMBOL(sock_set_rcvbuf);
1002 
1003 static void __sock_set_mark(struct sock *sk, u32 val)
1004 {
1005 	if (val != sk->sk_mark) {
1006 		WRITE_ONCE(sk->sk_mark, val);
1007 		sk_dst_reset(sk);
1008 	}
1009 }
1010 
1011 void sock_set_mark(struct sock *sk, u32 val)
1012 {
1013 	lock_sock(sk);
1014 	__sock_set_mark(sk, val);
1015 	release_sock(sk);
1016 }
1017 EXPORT_SYMBOL(sock_set_mark);
1018 
1019 static void sock_release_reserved_memory(struct sock *sk, int bytes)
1020 {
1021 	/* Round down bytes to multiple of pages */
1022 	bytes = round_down(bytes, PAGE_SIZE);
1023 
1024 	WARN_ON(bytes > sk->sk_reserved_mem);
1025 	WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes);
1026 	sk_mem_reclaim(sk);
1027 }
1028 
1029 static int sock_reserve_memory(struct sock *sk, int bytes)
1030 {
1031 	long allocated;
1032 	bool charged;
1033 	int pages;
1034 
1035 	if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk))
1036 		return -EOPNOTSUPP;
1037 
1038 	if (!bytes)
1039 		return 0;
1040 
1041 	pages = sk_mem_pages(bytes);
1042 
1043 	/* pre-charge to memcg */
1044 	charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages,
1045 					  GFP_KERNEL | __GFP_RETRY_MAYFAIL);
1046 	if (!charged)
1047 		return -ENOMEM;
1048 
1049 	/* pre-charge to forward_alloc */
1050 	sk_memory_allocated_add(sk, pages);
1051 	allocated = sk_memory_allocated(sk);
1052 	/* If the system goes into memory pressure with this
1053 	 * precharge, give up and return error.
1054 	 */
1055 	if (allocated > sk_prot_mem_limits(sk, 1)) {
1056 		sk_memory_allocated_sub(sk, pages);
1057 		mem_cgroup_uncharge_skmem(sk->sk_memcg, pages);
1058 		return -ENOMEM;
1059 	}
1060 	sk_forward_alloc_add(sk, pages << PAGE_SHIFT);
1061 
1062 	WRITE_ONCE(sk->sk_reserved_mem,
1063 		   sk->sk_reserved_mem + (pages << PAGE_SHIFT));
1064 
1065 	return 0;
1066 }
1067 
1068 #ifdef CONFIG_PAGE_POOL
1069 
1070 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED
1071  * in 1 syscall. The limit exists to limit the amount of memory the kernel
1072  * allocates to copy these tokens, and to prevent looping over the frags for
1073  * too long.
1074  */
1075 #define MAX_DONTNEED_TOKENS 128
1076 #define MAX_DONTNEED_FRAGS 1024
1077 
1078 static noinline_for_stack int
1079 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen)
1080 {
1081 	unsigned int num_tokens, i, j, k, netmem_num = 0;
1082 	struct dmabuf_token *tokens;
1083 	int ret = 0, num_frags = 0;
1084 	netmem_ref netmems[16];
1085 
1086 	if (!sk_is_tcp(sk))
1087 		return -EBADF;
1088 
1089 	if (optlen % sizeof(*tokens) ||
1090 	    optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS)
1091 		return -EINVAL;
1092 
1093 	num_tokens = optlen / sizeof(*tokens);
1094 	tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL);
1095 	if (!tokens)
1096 		return -ENOMEM;
1097 
1098 	if (copy_from_sockptr(tokens, optval, optlen)) {
1099 		kvfree(tokens);
1100 		return -EFAULT;
1101 	}
1102 
1103 	xa_lock_bh(&sk->sk_user_frags);
1104 	for (i = 0; i < num_tokens; i++) {
1105 		for (j = 0; j < tokens[i].token_count; j++) {
1106 			if (++num_frags > MAX_DONTNEED_FRAGS)
1107 				goto frag_limit_reached;
1108 
1109 			netmem_ref netmem = (__force netmem_ref)__xa_erase(
1110 				&sk->sk_user_frags, tokens[i].token_start + j);
1111 
1112 			if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem)))
1113 				continue;
1114 
1115 			netmems[netmem_num++] = netmem;
1116 			if (netmem_num == ARRAY_SIZE(netmems)) {
1117 				xa_unlock_bh(&sk->sk_user_frags);
1118 				for (k = 0; k < netmem_num; k++)
1119 					WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1120 				netmem_num = 0;
1121 				xa_lock_bh(&sk->sk_user_frags);
1122 			}
1123 			ret++;
1124 		}
1125 	}
1126 
1127 frag_limit_reached:
1128 	xa_unlock_bh(&sk->sk_user_frags);
1129 	for (k = 0; k < netmem_num; k++)
1130 		WARN_ON_ONCE(!napi_pp_put_page(netmems[k]));
1131 
1132 	kvfree(tokens);
1133 	return ret;
1134 }
1135 #endif
1136 
1137 void sockopt_lock_sock(struct sock *sk)
1138 {
1139 	/* When current->bpf_ctx is set, the setsockopt is called from
1140 	 * a bpf prog.  bpf has ensured the sk lock has been
1141 	 * acquired before calling setsockopt().
1142 	 */
1143 	if (has_current_bpf_ctx())
1144 		return;
1145 
1146 	lock_sock(sk);
1147 }
1148 EXPORT_SYMBOL(sockopt_lock_sock);
1149 
1150 void sockopt_release_sock(struct sock *sk)
1151 {
1152 	if (has_current_bpf_ctx())
1153 		return;
1154 
1155 	release_sock(sk);
1156 }
1157 EXPORT_SYMBOL(sockopt_release_sock);
1158 
1159 bool sockopt_ns_capable(struct user_namespace *ns, int cap)
1160 {
1161 	return has_current_bpf_ctx() || ns_capable(ns, cap);
1162 }
1163 EXPORT_SYMBOL(sockopt_ns_capable);
1164 
1165 bool sockopt_capable(int cap)
1166 {
1167 	return has_current_bpf_ctx() || capable(cap);
1168 }
1169 EXPORT_SYMBOL(sockopt_capable);
1170 
1171 static int sockopt_validate_clockid(__kernel_clockid_t value)
1172 {
1173 	switch (value) {
1174 	case CLOCK_REALTIME:
1175 	case CLOCK_MONOTONIC:
1176 	case CLOCK_TAI:
1177 		return 0;
1178 	}
1179 	return -EINVAL;
1180 }
1181 
1182 /*
1183  *	This is meant for all protocols to use and covers goings on
1184  *	at the socket level. Everything here is generic.
1185  */
1186 
1187 int sk_setsockopt(struct sock *sk, int level, int optname,
1188 		  sockptr_t optval, unsigned int optlen)
1189 {
1190 	struct so_timestamping timestamping;
1191 	struct socket *sock = sk->sk_socket;
1192 	struct sock_txtime sk_txtime;
1193 	int val;
1194 	int valbool;
1195 	struct linger ling;
1196 	int ret = 0;
1197 
1198 	/*
1199 	 *	Options without arguments
1200 	 */
1201 
1202 	if (optname == SO_BINDTODEVICE)
1203 		return sock_setbindtodevice(sk, optval, optlen);
1204 
1205 	if (optlen < sizeof(int))
1206 		return -EINVAL;
1207 
1208 	if (copy_from_sockptr(&val, optval, sizeof(val)))
1209 		return -EFAULT;
1210 
1211 	valbool = val ? 1 : 0;
1212 
1213 	/* handle options which do not require locking the socket. */
1214 	switch (optname) {
1215 	case SO_PRIORITY:
1216 		if (sk_set_prio_allowed(sk, val)) {
1217 			sock_set_priority(sk, val);
1218 			return 0;
1219 		}
1220 		return -EPERM;
1221 	case SO_TYPE:
1222 	case SO_PROTOCOL:
1223 	case SO_DOMAIN:
1224 	case SO_ERROR:
1225 		return -ENOPROTOOPT;
1226 #ifdef CONFIG_NET_RX_BUSY_POLL
1227 	case SO_BUSY_POLL:
1228 		if (val < 0)
1229 			return -EINVAL;
1230 		WRITE_ONCE(sk->sk_ll_usec, val);
1231 		return 0;
1232 	case SO_PREFER_BUSY_POLL:
1233 		if (valbool && !sockopt_capable(CAP_NET_ADMIN))
1234 			return -EPERM;
1235 		WRITE_ONCE(sk->sk_prefer_busy_poll, valbool);
1236 		return 0;
1237 	case SO_BUSY_POLL_BUDGET:
1238 		if (val > READ_ONCE(sk->sk_busy_poll_budget) &&
1239 		    !sockopt_capable(CAP_NET_ADMIN))
1240 			return -EPERM;
1241 		if (val < 0 || val > U16_MAX)
1242 			return -EINVAL;
1243 		WRITE_ONCE(sk->sk_busy_poll_budget, val);
1244 		return 0;
1245 #endif
1246 	case SO_MAX_PACING_RATE:
1247 		{
1248 		unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val;
1249 		unsigned long pacing_rate;
1250 
1251 		if (sizeof(ulval) != sizeof(val) &&
1252 		    optlen >= sizeof(ulval) &&
1253 		    copy_from_sockptr(&ulval, optval, sizeof(ulval))) {
1254 			return -EFAULT;
1255 		}
1256 		if (ulval != ~0UL)
1257 			cmpxchg(&sk->sk_pacing_status,
1258 				SK_PACING_NONE,
1259 				SK_PACING_NEEDED);
1260 		/* Pairs with READ_ONCE() from sk_getsockopt() */
1261 		WRITE_ONCE(sk->sk_max_pacing_rate, ulval);
1262 		pacing_rate = READ_ONCE(sk->sk_pacing_rate);
1263 		if (ulval < pacing_rate)
1264 			WRITE_ONCE(sk->sk_pacing_rate, ulval);
1265 		return 0;
1266 		}
1267 	case SO_TXREHASH:
1268 		if (!sk_is_tcp(sk))
1269 			return -EOPNOTSUPP;
1270 		if (val < -1 || val > 1)
1271 			return -EINVAL;
1272 		if ((u8)val == SOCK_TXREHASH_DEFAULT)
1273 			val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash);
1274 		/* Paired with READ_ONCE() in tcp_rtx_synack()
1275 		 * and sk_getsockopt().
1276 		 */
1277 		WRITE_ONCE(sk->sk_txrehash, (u8)val);
1278 		return 0;
1279 	case SO_PEEK_OFF:
1280 		{
1281 		int (*set_peek_off)(struct sock *sk, int val);
1282 
1283 		set_peek_off = READ_ONCE(sock->ops)->set_peek_off;
1284 		if (set_peek_off)
1285 			ret = set_peek_off(sk, val);
1286 		else
1287 			ret = -EOPNOTSUPP;
1288 		return ret;
1289 		}
1290 #ifdef CONFIG_PAGE_POOL
1291 	case SO_DEVMEM_DONTNEED:
1292 		return sock_devmem_dontneed(sk, optval, optlen);
1293 #endif
1294 	case SO_SNDTIMEO_OLD:
1295 	case SO_SNDTIMEO_NEW:
1296 		return sock_set_timeout(&sk->sk_sndtimeo, optval,
1297 					optlen, optname == SO_SNDTIMEO_OLD);
1298 	case SO_RCVTIMEO_OLD:
1299 	case SO_RCVTIMEO_NEW:
1300 		return sock_set_timeout(&sk->sk_rcvtimeo, optval,
1301 					optlen, optname == SO_RCVTIMEO_OLD);
1302 	}
1303 
1304 	sockopt_lock_sock(sk);
1305 
1306 	switch (optname) {
1307 	case SO_DEBUG:
1308 		if (val && !sockopt_capable(CAP_NET_ADMIN))
1309 			ret = -EACCES;
1310 		else
1311 			sock_valbool_flag(sk, SOCK_DBG, valbool);
1312 		break;
1313 	case SO_REUSEADDR:
1314 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
1315 		break;
1316 	case SO_REUSEPORT:
1317 		if (valbool && !sk_is_inet(sk))
1318 			ret = -EOPNOTSUPP;
1319 		else
1320 			sk->sk_reuseport = valbool;
1321 		break;
1322 	case SO_DONTROUTE:
1323 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
1324 		sk_dst_reset(sk);
1325 		break;
1326 	case SO_BROADCAST:
1327 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
1328 		break;
1329 	case SO_SNDBUF:
1330 		/* Don't error on this BSD doesn't and if you think
1331 		 * about it this is right. Otherwise apps have to
1332 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1333 		 * are treated in BSD as hints
1334 		 */
1335 		val = min_t(u32, val, READ_ONCE(sysctl_wmem_max));
1336 set_sndbuf:
1337 		/* Ensure val * 2 fits into an int, to prevent max_t()
1338 		 * from treating it as a negative value.
1339 		 */
1340 		val = min_t(int, val, INT_MAX / 2);
1341 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1342 		WRITE_ONCE(sk->sk_sndbuf,
1343 			   max_t(int, val * 2, SOCK_MIN_SNDBUF));
1344 		/* Wake up sending tasks if we upped the value. */
1345 		sk->sk_write_space(sk);
1346 		break;
1347 
1348 	case SO_SNDBUFFORCE:
1349 		if (!sockopt_capable(CAP_NET_ADMIN)) {
1350 			ret = -EPERM;
1351 			break;
1352 		}
1353 
1354 		/* No negative values (to prevent underflow, as val will be
1355 		 * multiplied by 2).
1356 		 */
1357 		if (val < 0)
1358 			val = 0;
1359 		goto set_sndbuf;
1360 
1361 	case SO_RCVBUF:
1362 		/* Don't error on this BSD doesn't and if you think
1363 		 * about it this is right. Otherwise apps have to
1364 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
1365 		 * are treated in BSD as hints
1366 		 */
1367 		__sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max)));
1368 		break;
1369 
1370 	case SO_RCVBUFFORCE:
1371 		if (!sockopt_capable(CAP_NET_ADMIN)) {
1372 			ret = -EPERM;
1373 			break;
1374 		}
1375 
1376 		/* No negative values (to prevent underflow, as val will be
1377 		 * multiplied by 2).
1378 		 */
1379 		__sock_set_rcvbuf(sk, max(val, 0));
1380 		break;
1381 
1382 	case SO_KEEPALIVE:
1383 		if (sk->sk_prot->keepalive)
1384 			sk->sk_prot->keepalive(sk, valbool);
1385 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
1386 		break;
1387 
1388 	case SO_OOBINLINE:
1389 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
1390 		break;
1391 
1392 	case SO_NO_CHECK:
1393 		sk->sk_no_check_tx = valbool;
1394 		break;
1395 
1396 	case SO_LINGER:
1397 		if (optlen < sizeof(ling)) {
1398 			ret = -EINVAL;	/* 1003.1g */
1399 			break;
1400 		}
1401 		if (copy_from_sockptr(&ling, optval, sizeof(ling))) {
1402 			ret = -EFAULT;
1403 			break;
1404 		}
1405 		if (!ling.l_onoff) {
1406 			sock_reset_flag(sk, SOCK_LINGER);
1407 		} else {
1408 			unsigned long t_sec = ling.l_linger;
1409 
1410 			if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ)
1411 				WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT);
1412 			else
1413 				WRITE_ONCE(sk->sk_lingertime, t_sec * HZ);
1414 			sock_set_flag(sk, SOCK_LINGER);
1415 		}
1416 		break;
1417 
1418 	case SO_BSDCOMPAT:
1419 		break;
1420 
1421 	case SO_TIMESTAMP_OLD:
1422 	case SO_TIMESTAMP_NEW:
1423 	case SO_TIMESTAMPNS_OLD:
1424 	case SO_TIMESTAMPNS_NEW:
1425 		sock_set_timestamp(sk, optname, valbool);
1426 		break;
1427 
1428 	case SO_TIMESTAMPING_NEW:
1429 	case SO_TIMESTAMPING_OLD:
1430 		if (optlen == sizeof(timestamping)) {
1431 			if (copy_from_sockptr(&timestamping, optval,
1432 					      sizeof(timestamping))) {
1433 				ret = -EFAULT;
1434 				break;
1435 			}
1436 		} else {
1437 			memset(&timestamping, 0, sizeof(timestamping));
1438 			timestamping.flags = val;
1439 		}
1440 		ret = sock_set_timestamping(sk, optname, timestamping);
1441 		break;
1442 
1443 	case SO_RCVLOWAT:
1444 		{
1445 		int (*set_rcvlowat)(struct sock *sk, int val) = NULL;
1446 
1447 		if (val < 0)
1448 			val = INT_MAX;
1449 		if (sock)
1450 			set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat;
1451 		if (set_rcvlowat)
1452 			ret = set_rcvlowat(sk, val);
1453 		else
1454 			WRITE_ONCE(sk->sk_rcvlowat, val ? : 1);
1455 		break;
1456 		}
1457 	case SO_ATTACH_FILTER: {
1458 		struct sock_fprog fprog;
1459 
1460 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1461 		if (!ret)
1462 			ret = sk_attach_filter(&fprog, sk);
1463 		break;
1464 	}
1465 	case SO_ATTACH_BPF:
1466 		ret = -EINVAL;
1467 		if (optlen == sizeof(u32)) {
1468 			u32 ufd;
1469 
1470 			ret = -EFAULT;
1471 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1472 				break;
1473 
1474 			ret = sk_attach_bpf(ufd, sk);
1475 		}
1476 		break;
1477 
1478 	case SO_ATTACH_REUSEPORT_CBPF: {
1479 		struct sock_fprog fprog;
1480 
1481 		ret = copy_bpf_fprog_from_user(&fprog, optval, optlen);
1482 		if (!ret)
1483 			ret = sk_reuseport_attach_filter(&fprog, sk);
1484 		break;
1485 	}
1486 	case SO_ATTACH_REUSEPORT_EBPF:
1487 		ret = -EINVAL;
1488 		if (optlen == sizeof(u32)) {
1489 			u32 ufd;
1490 
1491 			ret = -EFAULT;
1492 			if (copy_from_sockptr(&ufd, optval, sizeof(ufd)))
1493 				break;
1494 
1495 			ret = sk_reuseport_attach_bpf(ufd, sk);
1496 		}
1497 		break;
1498 
1499 	case SO_DETACH_REUSEPORT_BPF:
1500 		ret = reuseport_detach_prog(sk);
1501 		break;
1502 
1503 	case SO_DETACH_FILTER:
1504 		ret = sk_detach_filter(sk);
1505 		break;
1506 
1507 	case SO_LOCK_FILTER:
1508 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
1509 			ret = -EPERM;
1510 		else
1511 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
1512 		break;
1513 
1514 	case SO_MARK:
1515 		if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
1516 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1517 			ret = -EPERM;
1518 			break;
1519 		}
1520 
1521 		__sock_set_mark(sk, val);
1522 		break;
1523 	case SO_RCVMARK:
1524 		sock_valbool_flag(sk, SOCK_RCVMARK, valbool);
1525 		break;
1526 
1527 	case SO_RCVPRIORITY:
1528 		sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool);
1529 		break;
1530 
1531 	case SO_RXQ_OVFL:
1532 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
1533 		break;
1534 
1535 	case SO_WIFI_STATUS:
1536 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1537 		break;
1538 
1539 	case SO_NOFCS:
1540 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1541 		break;
1542 
1543 	case SO_SELECT_ERR_QUEUE:
1544 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1545 		break;
1546 
1547 	case SO_PASSCRED:
1548 		if (sk_may_scm_recv(sk))
1549 			sk->sk_scm_credentials = valbool;
1550 		else
1551 			ret = -EOPNOTSUPP;
1552 		break;
1553 
1554 	case SO_PASSSEC:
1555 		if (IS_ENABLED(CONFIG_SECURITY_NETWORK) && sk_may_scm_recv(sk))
1556 			sk->sk_scm_security = valbool;
1557 		else
1558 			ret = -EOPNOTSUPP;
1559 		break;
1560 
1561 	case SO_PASSPIDFD:
1562 		if (sk_is_unix(sk))
1563 			sk->sk_scm_pidfd = valbool;
1564 		else
1565 			ret = -EOPNOTSUPP;
1566 		break;
1567 
1568 	case SO_PASSRIGHTS:
1569 		if (sk_is_unix(sk))
1570 			sk->sk_scm_rights = valbool;
1571 		else
1572 			ret = -EOPNOTSUPP;
1573 		break;
1574 
1575 	case SO_INCOMING_CPU:
1576 		reuseport_update_incoming_cpu(sk, val);
1577 		break;
1578 
1579 	case SO_CNX_ADVICE:
1580 		if (val == 1)
1581 			dst_negative_advice(sk);
1582 		break;
1583 
1584 	case SO_ZEROCOPY:
1585 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
1586 			if (!(sk_is_tcp(sk) ||
1587 			      (sk->sk_type == SOCK_DGRAM &&
1588 			       sk->sk_protocol == IPPROTO_UDP)))
1589 				ret = -EOPNOTSUPP;
1590 		} else if (sk->sk_family != PF_RDS) {
1591 			ret = -EOPNOTSUPP;
1592 		}
1593 		if (!ret) {
1594 			if (val < 0 || val > 1)
1595 				ret = -EINVAL;
1596 			else
1597 				sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1598 		}
1599 		break;
1600 
1601 	case SO_TXTIME:
1602 		if (optlen != sizeof(struct sock_txtime)) {
1603 			ret = -EINVAL;
1604 			break;
1605 		} else if (copy_from_sockptr(&sk_txtime, optval,
1606 			   sizeof(struct sock_txtime))) {
1607 			ret = -EFAULT;
1608 			break;
1609 		} else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) {
1610 			ret = -EINVAL;
1611 			break;
1612 		}
1613 		/* CLOCK_MONOTONIC is only used by sch_fq, and this packet
1614 		 * scheduler has enough safe guards.
1615 		 */
1616 		if (sk_txtime.clockid != CLOCK_MONOTONIC &&
1617 		    !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) {
1618 			ret = -EPERM;
1619 			break;
1620 		}
1621 
1622 		ret = sockopt_validate_clockid(sk_txtime.clockid);
1623 		if (ret)
1624 			break;
1625 
1626 		sock_valbool_flag(sk, SOCK_TXTIME, true);
1627 		sk->sk_clockid = sk_txtime.clockid;
1628 		sk->sk_txtime_deadline_mode =
1629 			!!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE);
1630 		sk->sk_txtime_report_errors =
1631 			!!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS);
1632 		break;
1633 
1634 	case SO_BINDTOIFINDEX:
1635 		ret = sock_bindtoindex_locked(sk, val);
1636 		break;
1637 
1638 	case SO_BUF_LOCK:
1639 		if (val & ~SOCK_BUF_LOCK_MASK) {
1640 			ret = -EINVAL;
1641 			break;
1642 		}
1643 		sk->sk_userlocks = val | (sk->sk_userlocks &
1644 					  ~SOCK_BUF_LOCK_MASK);
1645 		break;
1646 
1647 	case SO_RESERVE_MEM:
1648 	{
1649 		int delta;
1650 
1651 		if (val < 0) {
1652 			ret = -EINVAL;
1653 			break;
1654 		}
1655 
1656 		delta = val - sk->sk_reserved_mem;
1657 		if (delta < 0)
1658 			sock_release_reserved_memory(sk, -delta);
1659 		else
1660 			ret = sock_reserve_memory(sk, delta);
1661 		break;
1662 	}
1663 
1664 	default:
1665 		ret = -ENOPROTOOPT;
1666 		break;
1667 	}
1668 	sockopt_release_sock(sk);
1669 	return ret;
1670 }
1671 
1672 int sock_setsockopt(struct socket *sock, int level, int optname,
1673 		    sockptr_t optval, unsigned int optlen)
1674 {
1675 	return sk_setsockopt(sock->sk, level, optname,
1676 			     optval, optlen);
1677 }
1678 EXPORT_SYMBOL(sock_setsockopt);
1679 
1680 static const struct cred *sk_get_peer_cred(struct sock *sk)
1681 {
1682 	const struct cred *cred;
1683 
1684 	spin_lock(&sk->sk_peer_lock);
1685 	cred = get_cred(sk->sk_peer_cred);
1686 	spin_unlock(&sk->sk_peer_lock);
1687 
1688 	return cred;
1689 }
1690 
1691 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1692 			  struct ucred *ucred)
1693 {
1694 	ucred->pid = pid_vnr(pid);
1695 	ucred->uid = ucred->gid = -1;
1696 	if (cred) {
1697 		struct user_namespace *current_ns = current_user_ns();
1698 
1699 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1700 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1701 	}
1702 }
1703 
1704 static int groups_to_user(sockptr_t dst, const struct group_info *src)
1705 {
1706 	struct user_namespace *user_ns = current_user_ns();
1707 	int i;
1708 
1709 	for (i = 0; i < src->ngroups; i++) {
1710 		gid_t gid = from_kgid_munged(user_ns, src->gid[i]);
1711 
1712 		if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid)))
1713 			return -EFAULT;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 int sk_getsockopt(struct sock *sk, int level, int optname,
1720 		  sockptr_t optval, sockptr_t optlen)
1721 {
1722 	struct socket *sock = sk->sk_socket;
1723 
1724 	union {
1725 		int val;
1726 		u64 val64;
1727 		unsigned long ulval;
1728 		struct linger ling;
1729 		struct old_timeval32 tm32;
1730 		struct __kernel_old_timeval tm;
1731 		struct  __kernel_sock_timeval stm;
1732 		struct sock_txtime txtime;
1733 		struct so_timestamping timestamping;
1734 	} v;
1735 
1736 	int lv = sizeof(int);
1737 	int len;
1738 
1739 	if (copy_from_sockptr(&len, optlen, sizeof(int)))
1740 		return -EFAULT;
1741 	if (len < 0)
1742 		return -EINVAL;
1743 
1744 	memset(&v, 0, sizeof(v));
1745 
1746 	switch (optname) {
1747 	case SO_DEBUG:
1748 		v.val = sock_flag(sk, SOCK_DBG);
1749 		break;
1750 
1751 	case SO_DONTROUTE:
1752 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1753 		break;
1754 
1755 	case SO_BROADCAST:
1756 		v.val = sock_flag(sk, SOCK_BROADCAST);
1757 		break;
1758 
1759 	case SO_SNDBUF:
1760 		v.val = READ_ONCE(sk->sk_sndbuf);
1761 		break;
1762 
1763 	case SO_RCVBUF:
1764 		v.val = READ_ONCE(sk->sk_rcvbuf);
1765 		break;
1766 
1767 	case SO_REUSEADDR:
1768 		v.val = sk->sk_reuse;
1769 		break;
1770 
1771 	case SO_REUSEPORT:
1772 		v.val = sk->sk_reuseport;
1773 		break;
1774 
1775 	case SO_KEEPALIVE:
1776 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1777 		break;
1778 
1779 	case SO_TYPE:
1780 		v.val = sk->sk_type;
1781 		break;
1782 
1783 	case SO_PROTOCOL:
1784 		v.val = sk->sk_protocol;
1785 		break;
1786 
1787 	case SO_DOMAIN:
1788 		v.val = sk->sk_family;
1789 		break;
1790 
1791 	case SO_ERROR:
1792 		v.val = -sock_error(sk);
1793 		if (v.val == 0)
1794 			v.val = xchg(&sk->sk_err_soft, 0);
1795 		break;
1796 
1797 	case SO_OOBINLINE:
1798 		v.val = sock_flag(sk, SOCK_URGINLINE);
1799 		break;
1800 
1801 	case SO_NO_CHECK:
1802 		v.val = sk->sk_no_check_tx;
1803 		break;
1804 
1805 	case SO_PRIORITY:
1806 		v.val = READ_ONCE(sk->sk_priority);
1807 		break;
1808 
1809 	case SO_LINGER:
1810 		lv		= sizeof(v.ling);
1811 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1812 		v.ling.l_linger	= READ_ONCE(sk->sk_lingertime) / HZ;
1813 		break;
1814 
1815 	case SO_BSDCOMPAT:
1816 		break;
1817 
1818 	case SO_TIMESTAMP_OLD:
1819 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1820 				!sock_flag(sk, SOCK_TSTAMP_NEW) &&
1821 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1822 		break;
1823 
1824 	case SO_TIMESTAMPNS_OLD:
1825 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW);
1826 		break;
1827 
1828 	case SO_TIMESTAMP_NEW:
1829 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW);
1830 		break;
1831 
1832 	case SO_TIMESTAMPNS_NEW:
1833 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW);
1834 		break;
1835 
1836 	case SO_TIMESTAMPING_OLD:
1837 	case SO_TIMESTAMPING_NEW:
1838 		lv = sizeof(v.timestamping);
1839 		/* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only
1840 		 * returning the flags when they were set through the same option.
1841 		 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD.
1842 		 */
1843 		if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) {
1844 			v.timestamping.flags = READ_ONCE(sk->sk_tsflags);
1845 			v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc);
1846 		}
1847 		break;
1848 
1849 	case SO_RCVTIMEO_OLD:
1850 	case SO_RCVTIMEO_NEW:
1851 		lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v,
1852 				      SO_RCVTIMEO_OLD == optname);
1853 		break;
1854 
1855 	case SO_SNDTIMEO_OLD:
1856 	case SO_SNDTIMEO_NEW:
1857 		lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v,
1858 				      SO_SNDTIMEO_OLD == optname);
1859 		break;
1860 
1861 	case SO_RCVLOWAT:
1862 		v.val = READ_ONCE(sk->sk_rcvlowat);
1863 		break;
1864 
1865 	case SO_SNDLOWAT:
1866 		v.val = 1;
1867 		break;
1868 
1869 	case SO_PASSCRED:
1870 		if (!sk_may_scm_recv(sk))
1871 			return -EOPNOTSUPP;
1872 
1873 		v.val = sk->sk_scm_credentials;
1874 		break;
1875 
1876 	case SO_PASSPIDFD:
1877 		if (!sk_is_unix(sk))
1878 			return -EOPNOTSUPP;
1879 
1880 		v.val = sk->sk_scm_pidfd;
1881 		break;
1882 
1883 	case SO_PASSRIGHTS:
1884 		if (!sk_is_unix(sk))
1885 			return -EOPNOTSUPP;
1886 
1887 		v.val = sk->sk_scm_rights;
1888 		break;
1889 
1890 	case SO_PEERCRED:
1891 	{
1892 		struct ucred peercred;
1893 		if (len > sizeof(peercred))
1894 			len = sizeof(peercred);
1895 
1896 		spin_lock(&sk->sk_peer_lock);
1897 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1898 		spin_unlock(&sk->sk_peer_lock);
1899 
1900 		if (copy_to_sockptr(optval, &peercred, len))
1901 			return -EFAULT;
1902 		goto lenout;
1903 	}
1904 
1905 	case SO_PEERPIDFD:
1906 	{
1907 		struct pid *peer_pid;
1908 		struct file *pidfd_file = NULL;
1909 		unsigned int flags = 0;
1910 		int pidfd;
1911 
1912 		if (len > sizeof(pidfd))
1913 			len = sizeof(pidfd);
1914 
1915 		spin_lock(&sk->sk_peer_lock);
1916 		peer_pid = get_pid(sk->sk_peer_pid);
1917 		spin_unlock(&sk->sk_peer_lock);
1918 
1919 		if (!peer_pid)
1920 			return -ENODATA;
1921 
1922 		/* The use of PIDFD_STALE requires stashing of struct pid
1923 		 * on pidfs with pidfs_register_pid() and only AF_UNIX
1924 		 * were prepared for this.
1925 		 */
1926 		if (sk->sk_family == AF_UNIX)
1927 			flags = PIDFD_STALE;
1928 
1929 		pidfd = pidfd_prepare(peer_pid, flags, &pidfd_file);
1930 		put_pid(peer_pid);
1931 		if (pidfd < 0)
1932 			return pidfd;
1933 
1934 		if (copy_to_sockptr(optval, &pidfd, len) ||
1935 		    copy_to_sockptr(optlen, &len, sizeof(int))) {
1936 			put_unused_fd(pidfd);
1937 			fput(pidfd_file);
1938 
1939 			return -EFAULT;
1940 		}
1941 
1942 		fd_install(pidfd, pidfd_file);
1943 		return 0;
1944 	}
1945 
1946 	case SO_PEERGROUPS:
1947 	{
1948 		const struct cred *cred;
1949 		int ret, n;
1950 
1951 		cred = sk_get_peer_cred(sk);
1952 		if (!cred)
1953 			return -ENODATA;
1954 
1955 		n = cred->group_info->ngroups;
1956 		if (len < n * sizeof(gid_t)) {
1957 			len = n * sizeof(gid_t);
1958 			put_cred(cred);
1959 			return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE;
1960 		}
1961 		len = n * sizeof(gid_t);
1962 
1963 		ret = groups_to_user(optval, cred->group_info);
1964 		put_cred(cred);
1965 		if (ret)
1966 			return ret;
1967 		goto lenout;
1968 	}
1969 
1970 	case SO_PEERNAME:
1971 	{
1972 		struct sockaddr_storage address;
1973 
1974 		lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2);
1975 		if (lv < 0)
1976 			return -ENOTCONN;
1977 		if (lv < len)
1978 			return -EINVAL;
1979 		if (copy_to_sockptr(optval, &address, len))
1980 			return -EFAULT;
1981 		goto lenout;
1982 	}
1983 
1984 	/* Dubious BSD thing... Probably nobody even uses it, but
1985 	 * the UNIX standard wants it for whatever reason... -DaveM
1986 	 */
1987 	case SO_ACCEPTCONN:
1988 		v.val = sk->sk_state == TCP_LISTEN;
1989 		break;
1990 
1991 	case SO_PASSSEC:
1992 		if (!IS_ENABLED(CONFIG_SECURITY_NETWORK) || !sk_may_scm_recv(sk))
1993 			return -EOPNOTSUPP;
1994 
1995 		v.val = sk->sk_scm_security;
1996 		break;
1997 
1998 	case SO_PEERSEC:
1999 		return security_socket_getpeersec_stream(sock,
2000 							 optval, optlen, len);
2001 
2002 	case SO_MARK:
2003 		v.val = READ_ONCE(sk->sk_mark);
2004 		break;
2005 
2006 	case SO_RCVMARK:
2007 		v.val = sock_flag(sk, SOCK_RCVMARK);
2008 		break;
2009 
2010 	case SO_RCVPRIORITY:
2011 		v.val = sock_flag(sk, SOCK_RCVPRIORITY);
2012 		break;
2013 
2014 	case SO_RXQ_OVFL:
2015 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
2016 		break;
2017 
2018 	case SO_WIFI_STATUS:
2019 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
2020 		break;
2021 
2022 	case SO_PEEK_OFF:
2023 		if (!READ_ONCE(sock->ops)->set_peek_off)
2024 			return -EOPNOTSUPP;
2025 
2026 		v.val = READ_ONCE(sk->sk_peek_off);
2027 		break;
2028 	case SO_NOFCS:
2029 		v.val = sock_flag(sk, SOCK_NOFCS);
2030 		break;
2031 
2032 	case SO_BINDTODEVICE:
2033 		return sock_getbindtodevice(sk, optval, optlen, len);
2034 
2035 	case SO_GET_FILTER:
2036 		len = sk_get_filter(sk, optval, len);
2037 		if (len < 0)
2038 			return len;
2039 
2040 		goto lenout;
2041 
2042 	case SO_LOCK_FILTER:
2043 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
2044 		break;
2045 
2046 	case SO_BPF_EXTENSIONS:
2047 		v.val = bpf_tell_extensions();
2048 		break;
2049 
2050 	case SO_SELECT_ERR_QUEUE:
2051 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
2052 		break;
2053 
2054 #ifdef CONFIG_NET_RX_BUSY_POLL
2055 	case SO_BUSY_POLL:
2056 		v.val = READ_ONCE(sk->sk_ll_usec);
2057 		break;
2058 	case SO_PREFER_BUSY_POLL:
2059 		v.val = READ_ONCE(sk->sk_prefer_busy_poll);
2060 		break;
2061 #endif
2062 
2063 	case SO_MAX_PACING_RATE:
2064 		/* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */
2065 		if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) {
2066 			lv = sizeof(v.ulval);
2067 			v.ulval = READ_ONCE(sk->sk_max_pacing_rate);
2068 		} else {
2069 			/* 32bit version */
2070 			v.val = min_t(unsigned long, ~0U,
2071 				      READ_ONCE(sk->sk_max_pacing_rate));
2072 		}
2073 		break;
2074 
2075 	case SO_INCOMING_CPU:
2076 		v.val = READ_ONCE(sk->sk_incoming_cpu);
2077 		break;
2078 
2079 	case SO_MEMINFO:
2080 	{
2081 		u32 meminfo[SK_MEMINFO_VARS];
2082 
2083 		sk_get_meminfo(sk, meminfo);
2084 
2085 		len = min_t(unsigned int, len, sizeof(meminfo));
2086 		if (copy_to_sockptr(optval, &meminfo, len))
2087 			return -EFAULT;
2088 
2089 		goto lenout;
2090 	}
2091 
2092 #ifdef CONFIG_NET_RX_BUSY_POLL
2093 	case SO_INCOMING_NAPI_ID:
2094 		v.val = READ_ONCE(sk->sk_napi_id);
2095 
2096 		/* aggregate non-NAPI IDs down to 0 */
2097 		if (!napi_id_valid(v.val))
2098 			v.val = 0;
2099 
2100 		break;
2101 #endif
2102 
2103 	case SO_COOKIE:
2104 		lv = sizeof(u64);
2105 		if (len < lv)
2106 			return -EINVAL;
2107 		v.val64 = sock_gen_cookie(sk);
2108 		break;
2109 
2110 	case SO_ZEROCOPY:
2111 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
2112 		break;
2113 
2114 	case SO_TXTIME:
2115 		lv = sizeof(v.txtime);
2116 		v.txtime.clockid = sk->sk_clockid;
2117 		v.txtime.flags |= sk->sk_txtime_deadline_mode ?
2118 				  SOF_TXTIME_DEADLINE_MODE : 0;
2119 		v.txtime.flags |= sk->sk_txtime_report_errors ?
2120 				  SOF_TXTIME_REPORT_ERRORS : 0;
2121 		break;
2122 
2123 	case SO_BINDTOIFINDEX:
2124 		v.val = READ_ONCE(sk->sk_bound_dev_if);
2125 		break;
2126 
2127 	case SO_NETNS_COOKIE:
2128 		lv = sizeof(u64);
2129 		if (len != lv)
2130 			return -EINVAL;
2131 		v.val64 = sock_net(sk)->net_cookie;
2132 		break;
2133 
2134 	case SO_BUF_LOCK:
2135 		v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK;
2136 		break;
2137 
2138 	case SO_RESERVE_MEM:
2139 		v.val = READ_ONCE(sk->sk_reserved_mem);
2140 		break;
2141 
2142 	case SO_TXREHASH:
2143 		if (!sk_is_tcp(sk))
2144 			return -EOPNOTSUPP;
2145 
2146 		/* Paired with WRITE_ONCE() in sk_setsockopt() */
2147 		v.val = READ_ONCE(sk->sk_txrehash);
2148 		break;
2149 
2150 	default:
2151 		/* We implement the SO_SNDLOWAT etc to not be settable
2152 		 * (1003.1g 7).
2153 		 */
2154 		return -ENOPROTOOPT;
2155 	}
2156 
2157 	if (len > lv)
2158 		len = lv;
2159 	if (copy_to_sockptr(optval, &v, len))
2160 		return -EFAULT;
2161 lenout:
2162 	if (copy_to_sockptr(optlen, &len, sizeof(int)))
2163 		return -EFAULT;
2164 	return 0;
2165 }
2166 
2167 /*
2168  * Initialize an sk_lock.
2169  *
2170  * (We also register the sk_lock with the lock validator.)
2171  */
2172 static inline void sock_lock_init(struct sock *sk)
2173 {
2174 	sk_owner_clear(sk);
2175 
2176 	if (sk->sk_kern_sock)
2177 		sock_lock_init_class_and_name(
2178 			sk,
2179 			af_family_kern_slock_key_strings[sk->sk_family],
2180 			af_family_kern_slock_keys + sk->sk_family,
2181 			af_family_kern_key_strings[sk->sk_family],
2182 			af_family_kern_keys + sk->sk_family);
2183 	else
2184 		sock_lock_init_class_and_name(
2185 			sk,
2186 			af_family_slock_key_strings[sk->sk_family],
2187 			af_family_slock_keys + sk->sk_family,
2188 			af_family_key_strings[sk->sk_family],
2189 			af_family_keys + sk->sk_family);
2190 }
2191 
2192 /*
2193  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
2194  * even temporarily, because of RCU lookups. sk_node should also be left as is.
2195  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
2196  */
2197 static void sock_copy(struct sock *nsk, const struct sock *osk)
2198 {
2199 	const struct proto *prot = READ_ONCE(osk->sk_prot);
2200 #ifdef CONFIG_SECURITY_NETWORK
2201 	void *sptr = nsk->sk_security;
2202 #endif
2203 
2204 	/* If we move sk_tx_queue_mapping out of the private section,
2205 	 * we must check if sk_tx_queue_clear() is called after
2206 	 * sock_copy() in sk_clone_lock().
2207 	 */
2208 	BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) <
2209 		     offsetof(struct sock, sk_dontcopy_begin) ||
2210 		     offsetof(struct sock, sk_tx_queue_mapping) >=
2211 		     offsetof(struct sock, sk_dontcopy_end));
2212 
2213 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
2214 
2215 	unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
2216 		      prot->obj_size - offsetof(struct sock, sk_dontcopy_end),
2217 		      /* alloc is larger than struct, see sk_prot_alloc() */);
2218 
2219 #ifdef CONFIG_SECURITY_NETWORK
2220 	nsk->sk_security = sptr;
2221 	security_sk_clone(osk, nsk);
2222 #endif
2223 }
2224 
2225 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
2226 		int family)
2227 {
2228 	struct sock *sk;
2229 	struct kmem_cache *slab;
2230 
2231 	slab = prot->slab;
2232 	if (slab != NULL) {
2233 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
2234 		if (!sk)
2235 			return sk;
2236 		if (want_init_on_alloc(priority))
2237 			sk_prot_clear_nulls(sk, prot->obj_size);
2238 	} else
2239 		sk = kmalloc(prot->obj_size, priority);
2240 
2241 	if (sk != NULL) {
2242 		if (security_sk_alloc(sk, family, priority))
2243 			goto out_free;
2244 
2245 		if (!try_module_get(prot->owner))
2246 			goto out_free_sec;
2247 	}
2248 
2249 	return sk;
2250 
2251 out_free_sec:
2252 	security_sk_free(sk);
2253 out_free:
2254 	if (slab != NULL)
2255 		kmem_cache_free(slab, sk);
2256 	else
2257 		kfree(sk);
2258 	return NULL;
2259 }
2260 
2261 static void sk_prot_free(struct proto *prot, struct sock *sk)
2262 {
2263 	struct kmem_cache *slab;
2264 	struct module *owner;
2265 
2266 	owner = prot->owner;
2267 	slab = prot->slab;
2268 
2269 	cgroup_sk_free(&sk->sk_cgrp_data);
2270 	mem_cgroup_sk_free(sk);
2271 	security_sk_free(sk);
2272 
2273 	sk_owner_put(sk);
2274 
2275 	if (slab != NULL)
2276 		kmem_cache_free(slab, sk);
2277 	else
2278 		kfree(sk);
2279 	module_put(owner);
2280 }
2281 
2282 /**
2283  *	sk_alloc - All socket objects are allocated here
2284  *	@net: the applicable net namespace
2285  *	@family: protocol family
2286  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2287  *	@prot: struct proto associated with this new sock instance
2288  *	@kern: is this to be a kernel socket?
2289  */
2290 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
2291 		      struct proto *prot, int kern)
2292 {
2293 	struct sock *sk;
2294 
2295 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
2296 	if (sk) {
2297 		sk->sk_family = family;
2298 		/*
2299 		 * See comment in struct sock definition to understand
2300 		 * why we need sk_prot_creator -acme
2301 		 */
2302 		sk->sk_prot = sk->sk_prot_creator = prot;
2303 		sk->sk_kern_sock = kern;
2304 		sock_lock_init(sk);
2305 		sk->sk_net_refcnt = kern ? 0 : 1;
2306 		if (likely(sk->sk_net_refcnt)) {
2307 			get_net_track(net, &sk->ns_tracker, priority);
2308 			sock_inuse_add(net, 1);
2309 		} else {
2310 			net_passive_inc(net);
2311 			__netns_tracker_alloc(net, &sk->ns_tracker,
2312 					      false, priority);
2313 		}
2314 
2315 		sock_net_set(sk, net);
2316 		refcount_set(&sk->sk_wmem_alloc, 1);
2317 
2318 		mem_cgroup_sk_alloc(sk);
2319 		cgroup_sk_alloc(&sk->sk_cgrp_data);
2320 		sock_update_classid(&sk->sk_cgrp_data);
2321 		sock_update_netprioidx(&sk->sk_cgrp_data);
2322 		sk_tx_queue_clear(sk);
2323 	}
2324 
2325 	return sk;
2326 }
2327 EXPORT_SYMBOL(sk_alloc);
2328 
2329 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
2330  * grace period. This is the case for UDP sockets and TCP listeners.
2331  */
2332 static void __sk_destruct(struct rcu_head *head)
2333 {
2334 	struct sock *sk = container_of(head, struct sock, sk_rcu);
2335 	struct net *net = sock_net(sk);
2336 	struct sk_filter *filter;
2337 
2338 	if (sk->sk_destruct)
2339 		sk->sk_destruct(sk);
2340 
2341 	filter = rcu_dereference_check(sk->sk_filter,
2342 				       refcount_read(&sk->sk_wmem_alloc) == 0);
2343 	if (filter) {
2344 		sk_filter_uncharge(sk, filter);
2345 		RCU_INIT_POINTER(sk->sk_filter, NULL);
2346 	}
2347 
2348 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
2349 
2350 #ifdef CONFIG_BPF_SYSCALL
2351 	bpf_sk_storage_free(sk);
2352 #endif
2353 
2354 	if (atomic_read(&sk->sk_omem_alloc))
2355 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
2356 			 __func__, atomic_read(&sk->sk_omem_alloc));
2357 
2358 	if (sk->sk_frag.page) {
2359 		put_page(sk->sk_frag.page);
2360 		sk->sk_frag.page = NULL;
2361 	}
2362 
2363 	/* We do not need to acquire sk->sk_peer_lock, we are the last user. */
2364 	put_cred(sk->sk_peer_cred);
2365 	put_pid(sk->sk_peer_pid);
2366 
2367 	if (likely(sk->sk_net_refcnt)) {
2368 		put_net_track(net, &sk->ns_tracker);
2369 	} else {
2370 		__netns_tracker_free(net, &sk->ns_tracker, false);
2371 		net_passive_dec(net);
2372 	}
2373 	sk_prot_free(sk->sk_prot_creator, sk);
2374 }
2375 
2376 void sk_net_refcnt_upgrade(struct sock *sk)
2377 {
2378 	struct net *net = sock_net(sk);
2379 
2380 	WARN_ON_ONCE(sk->sk_net_refcnt);
2381 	__netns_tracker_free(net, &sk->ns_tracker, false);
2382 	net_passive_dec(net);
2383 	sk->sk_net_refcnt = 1;
2384 	get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
2385 	sock_inuse_add(net, 1);
2386 }
2387 EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade);
2388 
2389 void sk_destruct(struct sock *sk)
2390 {
2391 	bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE);
2392 
2393 	if (rcu_access_pointer(sk->sk_reuseport_cb)) {
2394 		reuseport_detach_sock(sk);
2395 		use_call_rcu = true;
2396 	}
2397 
2398 	if (use_call_rcu)
2399 		call_rcu(&sk->sk_rcu, __sk_destruct);
2400 	else
2401 		__sk_destruct(&sk->sk_rcu);
2402 }
2403 
2404 static void __sk_free(struct sock *sk)
2405 {
2406 	if (likely(sk->sk_net_refcnt))
2407 		sock_inuse_add(sock_net(sk), -1);
2408 
2409 	if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk)))
2410 		sock_diag_broadcast_destroy(sk);
2411 	else
2412 		sk_destruct(sk);
2413 }
2414 
2415 void sk_free(struct sock *sk)
2416 {
2417 	/*
2418 	 * We subtract one from sk_wmem_alloc and can know if
2419 	 * some packets are still in some tx queue.
2420 	 * If not null, sock_wfree() will call __sk_free(sk) later
2421 	 */
2422 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
2423 		__sk_free(sk);
2424 }
2425 EXPORT_SYMBOL(sk_free);
2426 
2427 static void sk_init_common(struct sock *sk)
2428 {
2429 	skb_queue_head_init(&sk->sk_receive_queue);
2430 	skb_queue_head_init(&sk->sk_write_queue);
2431 	skb_queue_head_init(&sk->sk_error_queue);
2432 
2433 	rwlock_init(&sk->sk_callback_lock);
2434 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
2435 			af_rlock_keys + sk->sk_family,
2436 			af_family_rlock_key_strings[sk->sk_family]);
2437 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
2438 			af_wlock_keys + sk->sk_family,
2439 			af_family_wlock_key_strings[sk->sk_family]);
2440 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
2441 			af_elock_keys + sk->sk_family,
2442 			af_family_elock_key_strings[sk->sk_family]);
2443 	if (sk->sk_kern_sock)
2444 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2445 			af_kern_callback_keys + sk->sk_family,
2446 			af_family_kern_clock_key_strings[sk->sk_family]);
2447 	else
2448 		lockdep_set_class_and_name(&sk->sk_callback_lock,
2449 			af_callback_keys + sk->sk_family,
2450 			af_family_clock_key_strings[sk->sk_family]);
2451 }
2452 
2453 /**
2454  *	sk_clone_lock - clone a socket, and lock its clone
2455  *	@sk: the socket to clone
2456  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
2457  *
2458  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
2459  */
2460 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
2461 {
2462 	struct proto *prot = READ_ONCE(sk->sk_prot);
2463 	struct sk_filter *filter;
2464 	bool is_charged = true;
2465 	struct sock *newsk;
2466 
2467 	newsk = sk_prot_alloc(prot, priority, sk->sk_family);
2468 	if (!newsk)
2469 		goto out;
2470 
2471 	sock_copy(newsk, sk);
2472 
2473 	newsk->sk_prot_creator = prot;
2474 
2475 	/* SANITY */
2476 	if (likely(newsk->sk_net_refcnt)) {
2477 		get_net_track(sock_net(newsk), &newsk->ns_tracker, priority);
2478 		sock_inuse_add(sock_net(newsk), 1);
2479 	} else {
2480 		/* Kernel sockets are not elevating the struct net refcount.
2481 		 * Instead, use a tracker to more easily detect if a layer
2482 		 * is not properly dismantling its kernel sockets at netns
2483 		 * destroy time.
2484 		 */
2485 		net_passive_inc(sock_net(newsk));
2486 		__netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker,
2487 				      false, priority);
2488 	}
2489 	sk_node_init(&newsk->sk_node);
2490 	sock_lock_init(newsk);
2491 	bh_lock_sock(newsk);
2492 	newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
2493 	newsk->sk_backlog.len = 0;
2494 
2495 	atomic_set(&newsk->sk_rmem_alloc, 0);
2496 
2497 	/* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */
2498 	refcount_set(&newsk->sk_wmem_alloc, 1);
2499 
2500 	atomic_set(&newsk->sk_omem_alloc, 0);
2501 	sk_init_common(newsk);
2502 
2503 	newsk->sk_dst_cache	= NULL;
2504 	newsk->sk_dst_pending_confirm = 0;
2505 	newsk->sk_wmem_queued	= 0;
2506 	newsk->sk_forward_alloc = 0;
2507 	newsk->sk_reserved_mem  = 0;
2508 	atomic_set(&newsk->sk_drops, 0);
2509 	newsk->sk_send_head	= NULL;
2510 	newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
2511 	atomic_set(&newsk->sk_zckey, 0);
2512 
2513 	sock_reset_flag(newsk, SOCK_DONE);
2514 
2515 	/* sk->sk_memcg will be populated at accept() time */
2516 	newsk->sk_memcg = NULL;
2517 
2518 	cgroup_sk_clone(&newsk->sk_cgrp_data);
2519 
2520 	rcu_read_lock();
2521 	filter = rcu_dereference(sk->sk_filter);
2522 	if (filter != NULL)
2523 		/* though it's an empty new sock, the charging may fail
2524 		 * if sysctl_optmem_max was changed between creation of
2525 		 * original socket and cloning
2526 		 */
2527 		is_charged = sk_filter_charge(newsk, filter);
2528 	RCU_INIT_POINTER(newsk->sk_filter, filter);
2529 	rcu_read_unlock();
2530 
2531 	if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
2532 		/* We need to make sure that we don't uncharge the new
2533 		 * socket if we couldn't charge it in the first place
2534 		 * as otherwise we uncharge the parent's filter.
2535 		 */
2536 		if (!is_charged)
2537 			RCU_INIT_POINTER(newsk->sk_filter, NULL);
2538 
2539 		goto free;
2540 	}
2541 
2542 	RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
2543 
2544 	if (bpf_sk_storage_clone(sk, newsk))
2545 		goto free;
2546 
2547 	/* Clear sk_user_data if parent had the pointer tagged
2548 	 * as not suitable for copying when cloning.
2549 	 */
2550 	if (sk_user_data_is_nocopy(newsk))
2551 		newsk->sk_user_data = NULL;
2552 
2553 	newsk->sk_err	   = 0;
2554 	newsk->sk_err_soft = 0;
2555 	newsk->sk_priority = 0;
2556 	newsk->sk_incoming_cpu = raw_smp_processor_id();
2557 
2558 	/* Before updating sk_refcnt, we must commit prior changes to memory
2559 	 * (Documentation/RCU/rculist_nulls.rst for details)
2560 	 */
2561 	smp_wmb();
2562 	refcount_set(&newsk->sk_refcnt, 2);
2563 
2564 	sk_set_socket(newsk, NULL);
2565 	sk_tx_queue_clear(newsk);
2566 	RCU_INIT_POINTER(newsk->sk_wq, NULL);
2567 
2568 	if (newsk->sk_prot->sockets_allocated)
2569 		sk_sockets_allocated_inc(newsk);
2570 
2571 	if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP)
2572 		net_enable_timestamp();
2573 out:
2574 	return newsk;
2575 free:
2576 	/* It is still raw copy of parent, so invalidate
2577 	 * destructor and make plain sk_free()
2578 	 */
2579 	newsk->sk_destruct = NULL;
2580 	bh_unlock_sock(newsk);
2581 	sk_free(newsk);
2582 	newsk = NULL;
2583 	goto out;
2584 }
2585 EXPORT_SYMBOL_GPL(sk_clone_lock);
2586 
2587 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst)
2588 {
2589 	bool is_ipv6 = false;
2590 	u32 max_size;
2591 
2592 #if IS_ENABLED(CONFIG_IPV6)
2593 	is_ipv6 = (sk->sk_family == AF_INET6 &&
2594 		   !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr));
2595 #endif
2596 	/* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */
2597 	max_size = is_ipv6 ? READ_ONCE(dst_dev(dst)->gso_max_size) :
2598 			READ_ONCE(dst_dev(dst)->gso_ipv4_max_size);
2599 	if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk))
2600 		max_size = GSO_LEGACY_MAX_SIZE;
2601 
2602 	return max_size - (MAX_TCP_HEADER + 1);
2603 }
2604 
2605 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
2606 {
2607 	u32 max_segs = 1;
2608 
2609 	sk->sk_route_caps = dst_dev(dst)->features;
2610 	if (sk_is_tcp(sk)) {
2611 		struct inet_connection_sock *icsk = inet_csk(sk);
2612 
2613 		sk->sk_route_caps |= NETIF_F_GSO;
2614 		icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK);
2615 	}
2616 	if (sk->sk_route_caps & NETIF_F_GSO)
2617 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
2618 	if (unlikely(sk->sk_gso_disabled))
2619 		sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2620 	if (sk_can_gso(sk)) {
2621 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
2622 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2623 		} else {
2624 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
2625 			sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst);
2626 			/* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */
2627 			max_segs = max_t(u32, READ_ONCE(dst_dev(dst)->gso_max_segs), 1);
2628 		}
2629 	}
2630 	sk->sk_gso_max_segs = max_segs;
2631 	sk_dst_set(sk, dst);
2632 }
2633 EXPORT_SYMBOL_GPL(sk_setup_caps);
2634 
2635 /*
2636  *	Simple resource managers for sockets.
2637  */
2638 
2639 
2640 /*
2641  * Write buffer destructor automatically called from kfree_skb.
2642  */
2643 void sock_wfree(struct sk_buff *skb)
2644 {
2645 	struct sock *sk = skb->sk;
2646 	unsigned int len = skb->truesize;
2647 	bool free;
2648 
2649 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
2650 		if (sock_flag(sk, SOCK_RCU_FREE) &&
2651 		    sk->sk_write_space == sock_def_write_space) {
2652 			rcu_read_lock();
2653 			free = refcount_sub_and_test(len, &sk->sk_wmem_alloc);
2654 			sock_def_write_space_wfree(sk);
2655 			rcu_read_unlock();
2656 			if (unlikely(free))
2657 				__sk_free(sk);
2658 			return;
2659 		}
2660 
2661 		/*
2662 		 * Keep a reference on sk_wmem_alloc, this will be released
2663 		 * after sk_write_space() call
2664 		 */
2665 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
2666 		sk->sk_write_space(sk);
2667 		len = 1;
2668 	}
2669 	/*
2670 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
2671 	 * could not do because of in-flight packets
2672 	 */
2673 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
2674 		__sk_free(sk);
2675 }
2676 EXPORT_SYMBOL(sock_wfree);
2677 
2678 /* This variant of sock_wfree() is used by TCP,
2679  * since it sets SOCK_USE_WRITE_QUEUE.
2680  */
2681 void __sock_wfree(struct sk_buff *skb)
2682 {
2683 	struct sock *sk = skb->sk;
2684 
2685 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
2686 		__sk_free(sk);
2687 }
2688 
2689 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2690 {
2691 	skb_orphan(skb);
2692 #ifdef CONFIG_INET
2693 	if (unlikely(!sk_fullsock(sk)))
2694 		return skb_set_owner_edemux(skb, sk);
2695 #endif
2696 	skb->sk = sk;
2697 	skb->destructor = sock_wfree;
2698 	skb_set_hash_from_sk(skb, sk);
2699 	/*
2700 	 * We used to take a refcount on sk, but following operation
2701 	 * is enough to guarantee sk_free() won't free this sock until
2702 	 * all in-flight packets are completed
2703 	 */
2704 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
2705 }
2706 EXPORT_SYMBOL(skb_set_owner_w);
2707 
2708 static bool can_skb_orphan_partial(const struct sk_buff *skb)
2709 {
2710 	/* Drivers depend on in-order delivery for crypto offload,
2711 	 * partial orphan breaks out-of-order-OK logic.
2712 	 */
2713 	if (skb_is_decrypted(skb))
2714 		return false;
2715 
2716 	return (skb->destructor == sock_wfree ||
2717 		(IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree));
2718 }
2719 
2720 /* This helper is used by netem, as it can hold packets in its
2721  * delay queue. We want to allow the owner socket to send more
2722  * packets, as if they were already TX completed by a typical driver.
2723  * But we also want to keep skb->sk set because some packet schedulers
2724  * rely on it (sch_fq for example).
2725  */
2726 void skb_orphan_partial(struct sk_buff *skb)
2727 {
2728 	if (skb_is_tcp_pure_ack(skb))
2729 		return;
2730 
2731 	if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk))
2732 		return;
2733 
2734 	skb_orphan(skb);
2735 }
2736 EXPORT_SYMBOL(skb_orphan_partial);
2737 
2738 /*
2739  * Read buffer destructor automatically called from kfree_skb.
2740  */
2741 void sock_rfree(struct sk_buff *skb)
2742 {
2743 	struct sock *sk = skb->sk;
2744 	unsigned int len = skb->truesize;
2745 
2746 	atomic_sub(len, &sk->sk_rmem_alloc);
2747 	sk_mem_uncharge(sk, len);
2748 }
2749 EXPORT_SYMBOL(sock_rfree);
2750 
2751 /*
2752  * Buffer destructor for skbs that are not used directly in read or write
2753  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
2754  */
2755 void sock_efree(struct sk_buff *skb)
2756 {
2757 	sock_put(skb->sk);
2758 }
2759 EXPORT_SYMBOL(sock_efree);
2760 
2761 /* Buffer destructor for prefetch/receive path where reference count may
2762  * not be held, e.g. for listen sockets.
2763  */
2764 #ifdef CONFIG_INET
2765 void sock_pfree(struct sk_buff *skb)
2766 {
2767 	struct sock *sk = skb->sk;
2768 
2769 	if (!sk_is_refcounted(sk))
2770 		return;
2771 
2772 	if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) {
2773 		inet_reqsk(sk)->rsk_listener = NULL;
2774 		reqsk_free(inet_reqsk(sk));
2775 		return;
2776 	}
2777 
2778 	sock_gen_put(sk);
2779 }
2780 EXPORT_SYMBOL(sock_pfree);
2781 #endif /* CONFIG_INET */
2782 
2783 /*
2784  * Allocate a skb from the socket's send buffer.
2785  */
2786 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
2787 			     gfp_t priority)
2788 {
2789 	if (force ||
2790 	    refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) {
2791 		struct sk_buff *skb = alloc_skb(size, priority);
2792 
2793 		if (skb) {
2794 			skb_set_owner_w(skb, sk);
2795 			return skb;
2796 		}
2797 	}
2798 	return NULL;
2799 }
2800 EXPORT_SYMBOL(sock_wmalloc);
2801 
2802 static void sock_ofree(struct sk_buff *skb)
2803 {
2804 	struct sock *sk = skb->sk;
2805 
2806 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
2807 }
2808 
2809 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
2810 			     gfp_t priority)
2811 {
2812 	struct sk_buff *skb;
2813 
2814 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
2815 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
2816 	    READ_ONCE(sock_net(sk)->core.sysctl_optmem_max))
2817 		return NULL;
2818 
2819 	skb = alloc_skb(size, priority);
2820 	if (!skb)
2821 		return NULL;
2822 
2823 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
2824 	skb->sk = sk;
2825 	skb->destructor = sock_ofree;
2826 	return skb;
2827 }
2828 
2829 /*
2830  * Allocate a memory block from the socket's option memory buffer.
2831  */
2832 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
2833 {
2834 	int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max);
2835 
2836 	if ((unsigned int)size <= optmem_max &&
2837 	    atomic_read(&sk->sk_omem_alloc) + size < optmem_max) {
2838 		void *mem;
2839 		/* First do the add, to avoid the race if kmalloc
2840 		 * might sleep.
2841 		 */
2842 		atomic_add(size, &sk->sk_omem_alloc);
2843 		mem = kmalloc(size, priority);
2844 		if (mem)
2845 			return mem;
2846 		atomic_sub(size, &sk->sk_omem_alloc);
2847 	}
2848 	return NULL;
2849 }
2850 EXPORT_SYMBOL(sock_kmalloc);
2851 
2852 /*
2853  * Duplicate the input "src" memory block using the socket's
2854  * option memory buffer.
2855  */
2856 void *sock_kmemdup(struct sock *sk, const void *src,
2857 		   int size, gfp_t priority)
2858 {
2859 	void *mem;
2860 
2861 	mem = sock_kmalloc(sk, size, priority);
2862 	if (mem)
2863 		memcpy(mem, src, size);
2864 	return mem;
2865 }
2866 EXPORT_SYMBOL(sock_kmemdup);
2867 
2868 /* Free an option memory block. Note, we actually want the inline
2869  * here as this allows gcc to detect the nullify and fold away the
2870  * condition entirely.
2871  */
2872 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2873 				  const bool nullify)
2874 {
2875 	if (WARN_ON_ONCE(!mem))
2876 		return;
2877 	if (nullify)
2878 		kfree_sensitive(mem);
2879 	else
2880 		kfree(mem);
2881 	atomic_sub(size, &sk->sk_omem_alloc);
2882 }
2883 
2884 void sock_kfree_s(struct sock *sk, void *mem, int size)
2885 {
2886 	__sock_kfree_s(sk, mem, size, false);
2887 }
2888 EXPORT_SYMBOL(sock_kfree_s);
2889 
2890 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2891 {
2892 	__sock_kfree_s(sk, mem, size, true);
2893 }
2894 EXPORT_SYMBOL(sock_kzfree_s);
2895 
2896 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2897    I think, these locks should be removed for datagram sockets.
2898  */
2899 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2900 {
2901 	DEFINE_WAIT(wait);
2902 
2903 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2904 	for (;;) {
2905 		if (!timeo)
2906 			break;
2907 		if (signal_pending(current))
2908 			break;
2909 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2910 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2911 		if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf))
2912 			break;
2913 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2914 			break;
2915 		if (READ_ONCE(sk->sk_err))
2916 			break;
2917 		timeo = schedule_timeout(timeo);
2918 	}
2919 	finish_wait(sk_sleep(sk), &wait);
2920 	return timeo;
2921 }
2922 
2923 
2924 /*
2925  *	Generic send/receive buffer handlers
2926  */
2927 
2928 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2929 				     unsigned long data_len, int noblock,
2930 				     int *errcode, int max_page_order)
2931 {
2932 	struct sk_buff *skb;
2933 	long timeo;
2934 	int err;
2935 
2936 	timeo = sock_sndtimeo(sk, noblock);
2937 	for (;;) {
2938 		err = sock_error(sk);
2939 		if (err != 0)
2940 			goto failure;
2941 
2942 		err = -EPIPE;
2943 		if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN)
2944 			goto failure;
2945 
2946 		if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf))
2947 			break;
2948 
2949 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2950 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2951 		err = -EAGAIN;
2952 		if (!timeo)
2953 			goto failure;
2954 		if (signal_pending(current))
2955 			goto interrupted;
2956 		timeo = sock_wait_for_wmem(sk, timeo);
2957 	}
2958 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2959 				   errcode, sk->sk_allocation);
2960 	if (skb)
2961 		skb_set_owner_w(skb, sk);
2962 	return skb;
2963 
2964 interrupted:
2965 	err = sock_intr_errno(timeo);
2966 failure:
2967 	*errcode = err;
2968 	return NULL;
2969 }
2970 EXPORT_SYMBOL(sock_alloc_send_pskb);
2971 
2972 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
2973 		     struct sockcm_cookie *sockc)
2974 {
2975 	u32 tsflags;
2976 
2977 	BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31));
2978 
2979 	switch (cmsg->cmsg_type) {
2980 	case SO_MARK:
2981 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) &&
2982 		    !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2983 			return -EPERM;
2984 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2985 			return -EINVAL;
2986 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2987 		break;
2988 	case SO_TIMESTAMPING_OLD:
2989 	case SO_TIMESTAMPING_NEW:
2990 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2991 			return -EINVAL;
2992 
2993 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2994 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2995 			return -EINVAL;
2996 
2997 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2998 		sockc->tsflags |= tsflags;
2999 		break;
3000 	case SCM_TXTIME:
3001 		if (!sock_flag(sk, SOCK_TXTIME))
3002 			return -EINVAL;
3003 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64)))
3004 			return -EINVAL;
3005 		sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg));
3006 		break;
3007 	case SCM_TS_OPT_ID:
3008 		if (sk_is_tcp(sk))
3009 			return -EINVAL;
3010 		tsflags = READ_ONCE(sk->sk_tsflags);
3011 		if (!(tsflags & SOF_TIMESTAMPING_OPT_ID))
3012 			return -EINVAL;
3013 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3014 			return -EINVAL;
3015 		sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg);
3016 		sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID;
3017 		break;
3018 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
3019 	case SCM_RIGHTS:
3020 	case SCM_CREDENTIALS:
3021 		break;
3022 	case SO_PRIORITY:
3023 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3024 			return -EINVAL;
3025 		if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg)))
3026 			return -EPERM;
3027 		sockc->priority = *(u32 *)CMSG_DATA(cmsg);
3028 		break;
3029 	case SCM_DEVMEM_DMABUF:
3030 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
3031 			return -EINVAL;
3032 		sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg);
3033 		break;
3034 	default:
3035 		return -EINVAL;
3036 	}
3037 	return 0;
3038 }
3039 EXPORT_SYMBOL(__sock_cmsg_send);
3040 
3041 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
3042 		   struct sockcm_cookie *sockc)
3043 {
3044 	struct cmsghdr *cmsg;
3045 	int ret;
3046 
3047 	for_each_cmsghdr(cmsg, msg) {
3048 		if (!CMSG_OK(msg, cmsg))
3049 			return -EINVAL;
3050 		if (cmsg->cmsg_level != SOL_SOCKET)
3051 			continue;
3052 		ret = __sock_cmsg_send(sk, cmsg, sockc);
3053 		if (ret)
3054 			return ret;
3055 	}
3056 	return 0;
3057 }
3058 EXPORT_SYMBOL(sock_cmsg_send);
3059 
3060 static void sk_enter_memory_pressure(struct sock *sk)
3061 {
3062 	if (!sk->sk_prot->enter_memory_pressure)
3063 		return;
3064 
3065 	sk->sk_prot->enter_memory_pressure(sk);
3066 }
3067 
3068 static void sk_leave_memory_pressure(struct sock *sk)
3069 {
3070 	if (sk->sk_prot->leave_memory_pressure) {
3071 		INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure,
3072 				     tcp_leave_memory_pressure, sk);
3073 	} else {
3074 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
3075 
3076 		if (memory_pressure && READ_ONCE(*memory_pressure))
3077 			WRITE_ONCE(*memory_pressure, 0);
3078 	}
3079 }
3080 
3081 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3082 
3083 /**
3084  * skb_page_frag_refill - check that a page_frag contains enough room
3085  * @sz: minimum size of the fragment we want to get
3086  * @pfrag: pointer to page_frag
3087  * @gfp: priority for memory allocation
3088  *
3089  * Note: While this allocator tries to use high order pages, there is
3090  * no guarantee that allocations succeed. Therefore, @sz MUST be
3091  * less or equal than PAGE_SIZE.
3092  */
3093 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
3094 {
3095 	if (pfrag->page) {
3096 		if (page_ref_count(pfrag->page) == 1) {
3097 			pfrag->offset = 0;
3098 			return true;
3099 		}
3100 		if (pfrag->offset + sz <= pfrag->size)
3101 			return true;
3102 		put_page(pfrag->page);
3103 	}
3104 
3105 	pfrag->offset = 0;
3106 	if (SKB_FRAG_PAGE_ORDER &&
3107 	    !static_branch_unlikely(&net_high_order_alloc_disable_key)) {
3108 		/* Avoid direct reclaim but allow kswapd to wake */
3109 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
3110 					  __GFP_COMP | __GFP_NOWARN |
3111 					  __GFP_NORETRY,
3112 					  SKB_FRAG_PAGE_ORDER);
3113 		if (likely(pfrag->page)) {
3114 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
3115 			return true;
3116 		}
3117 	}
3118 	pfrag->page = alloc_page(gfp);
3119 	if (likely(pfrag->page)) {
3120 		pfrag->size = PAGE_SIZE;
3121 		return true;
3122 	}
3123 	return false;
3124 }
3125 EXPORT_SYMBOL(skb_page_frag_refill);
3126 
3127 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
3128 {
3129 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
3130 		return true;
3131 
3132 	sk_enter_memory_pressure(sk);
3133 	sk_stream_moderate_sndbuf(sk);
3134 	return false;
3135 }
3136 EXPORT_SYMBOL(sk_page_frag_refill);
3137 
3138 void __lock_sock(struct sock *sk)
3139 	__releases(&sk->sk_lock.slock)
3140 	__acquires(&sk->sk_lock.slock)
3141 {
3142 	DEFINE_WAIT(wait);
3143 
3144 	for (;;) {
3145 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
3146 					TASK_UNINTERRUPTIBLE);
3147 		spin_unlock_bh(&sk->sk_lock.slock);
3148 		schedule();
3149 		spin_lock_bh(&sk->sk_lock.slock);
3150 		if (!sock_owned_by_user(sk))
3151 			break;
3152 	}
3153 	finish_wait(&sk->sk_lock.wq, &wait);
3154 }
3155 
3156 void __release_sock(struct sock *sk)
3157 	__releases(&sk->sk_lock.slock)
3158 	__acquires(&sk->sk_lock.slock)
3159 {
3160 	struct sk_buff *skb, *next;
3161 
3162 	while ((skb = sk->sk_backlog.head) != NULL) {
3163 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
3164 
3165 		spin_unlock_bh(&sk->sk_lock.slock);
3166 
3167 		do {
3168 			next = skb->next;
3169 			prefetch(next);
3170 			DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb));
3171 			skb_mark_not_on_list(skb);
3172 			sk_backlog_rcv(sk, skb);
3173 
3174 			cond_resched();
3175 
3176 			skb = next;
3177 		} while (skb != NULL);
3178 
3179 		spin_lock_bh(&sk->sk_lock.slock);
3180 	}
3181 
3182 	/*
3183 	 * Doing the zeroing here guarantee we can not loop forever
3184 	 * while a wild producer attempts to flood us.
3185 	 */
3186 	sk->sk_backlog.len = 0;
3187 }
3188 
3189 void __sk_flush_backlog(struct sock *sk)
3190 {
3191 	spin_lock_bh(&sk->sk_lock.slock);
3192 	__release_sock(sk);
3193 
3194 	if (sk->sk_prot->release_cb)
3195 		INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3196 				     tcp_release_cb, sk);
3197 
3198 	spin_unlock_bh(&sk->sk_lock.slock);
3199 }
3200 EXPORT_SYMBOL_GPL(__sk_flush_backlog);
3201 
3202 /**
3203  * sk_wait_data - wait for data to arrive at sk_receive_queue
3204  * @sk:    sock to wait on
3205  * @timeo: for how long
3206  * @skb:   last skb seen on sk_receive_queue
3207  *
3208  * Now socket state including sk->sk_err is changed only under lock,
3209  * hence we may omit checks after joining wait queue.
3210  * We check receive queue before schedule() only as optimization;
3211  * it is very likely that release_sock() added new data.
3212  */
3213 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
3214 {
3215 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
3216 	int rc;
3217 
3218 	add_wait_queue(sk_sleep(sk), &wait);
3219 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3220 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
3221 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
3222 	remove_wait_queue(sk_sleep(sk), &wait);
3223 	return rc;
3224 }
3225 EXPORT_SYMBOL(sk_wait_data);
3226 
3227 /**
3228  *	__sk_mem_raise_allocated - increase memory_allocated
3229  *	@sk: socket
3230  *	@size: memory size to allocate
3231  *	@amt: pages to allocate
3232  *	@kind: allocation type
3233  *
3234  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc.
3235  *
3236  *	Unlike the globally shared limits among the sockets under same protocol,
3237  *	consuming the budget of a memcg won't have direct effect on other ones.
3238  *	So be optimistic about memcg's tolerance, and leave the callers to decide
3239  *	whether or not to raise allocated through sk_under_memory_pressure() or
3240  *	its variants.
3241  */
3242 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
3243 {
3244 	struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL;
3245 	struct proto *prot = sk->sk_prot;
3246 	bool charged = true;
3247 	long allocated;
3248 
3249 	sk_memory_allocated_add(sk, amt);
3250 	allocated = sk_memory_allocated(sk);
3251 
3252 	if (memcg) {
3253 		charged = mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge());
3254 		if (!charged)
3255 			goto suppress_allocation;
3256 	}
3257 
3258 	/* Under limit. */
3259 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
3260 		sk_leave_memory_pressure(sk);
3261 		return 1;
3262 	}
3263 
3264 	/* Under pressure. */
3265 	if (allocated > sk_prot_mem_limits(sk, 1))
3266 		sk_enter_memory_pressure(sk);
3267 
3268 	/* Over hard limit. */
3269 	if (allocated > sk_prot_mem_limits(sk, 2))
3270 		goto suppress_allocation;
3271 
3272 	/* Guarantee minimum buffer size under pressure (either global
3273 	 * or memcg) to make sure features described in RFC 7323 (TCP
3274 	 * Extensions for High Performance) work properly.
3275 	 *
3276 	 * This rule does NOT stand when exceeds global or memcg's hard
3277 	 * limit, or else a DoS attack can be taken place by spawning
3278 	 * lots of sockets whose usage are under minimum buffer size.
3279 	 */
3280 	if (kind == SK_MEM_RECV) {
3281 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
3282 			return 1;
3283 
3284 	} else { /* SK_MEM_SEND */
3285 		int wmem0 = sk_get_wmem0(sk, prot);
3286 
3287 		if (sk->sk_type == SOCK_STREAM) {
3288 			if (sk->sk_wmem_queued < wmem0)
3289 				return 1;
3290 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
3291 				return 1;
3292 		}
3293 	}
3294 
3295 	if (sk_has_memory_pressure(sk)) {
3296 		u64 alloc;
3297 
3298 		/* The following 'average' heuristic is within the
3299 		 * scope of global accounting, so it only makes
3300 		 * sense for global memory pressure.
3301 		 */
3302 		if (!sk_under_global_memory_pressure(sk))
3303 			return 1;
3304 
3305 		/* Try to be fair among all the sockets under global
3306 		 * pressure by allowing the ones that below average
3307 		 * usage to raise.
3308 		 */
3309 		alloc = sk_sockets_allocated_read_positive(sk);
3310 		if (sk_prot_mem_limits(sk, 2) > alloc *
3311 		    sk_mem_pages(sk->sk_wmem_queued +
3312 				 atomic_read(&sk->sk_rmem_alloc) +
3313 				 sk->sk_forward_alloc))
3314 			return 1;
3315 	}
3316 
3317 suppress_allocation:
3318 
3319 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
3320 		sk_stream_moderate_sndbuf(sk);
3321 
3322 		/* Fail only if socket is _under_ its sndbuf.
3323 		 * In this case we cannot block, so that we have to fail.
3324 		 */
3325 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) {
3326 			/* Force charge with __GFP_NOFAIL */
3327 			if (memcg && !charged) {
3328 				mem_cgroup_charge_skmem(memcg, amt,
3329 					gfp_memcg_charge() | __GFP_NOFAIL);
3330 			}
3331 			return 1;
3332 		}
3333 	}
3334 
3335 	if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged))
3336 		trace_sock_exceed_buf_limit(sk, prot, allocated, kind);
3337 
3338 	sk_memory_allocated_sub(sk, amt);
3339 
3340 	if (memcg && charged)
3341 		mem_cgroup_uncharge_skmem(memcg, amt);
3342 
3343 	return 0;
3344 }
3345 
3346 /**
3347  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
3348  *	@sk: socket
3349  *	@size: memory size to allocate
3350  *	@kind: allocation type
3351  *
3352  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
3353  *	rmem allocation. This function assumes that protocols which have
3354  *	memory_pressure use sk_wmem_queued as write buffer accounting.
3355  */
3356 int __sk_mem_schedule(struct sock *sk, int size, int kind)
3357 {
3358 	int ret, amt = sk_mem_pages(size);
3359 
3360 	sk_forward_alloc_add(sk, amt << PAGE_SHIFT);
3361 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
3362 	if (!ret)
3363 		sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT));
3364 	return ret;
3365 }
3366 EXPORT_SYMBOL(__sk_mem_schedule);
3367 
3368 /**
3369  *	__sk_mem_reduce_allocated - reclaim memory_allocated
3370  *	@sk: socket
3371  *	@amount: number of quanta
3372  *
3373  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
3374  */
3375 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
3376 {
3377 	sk_memory_allocated_sub(sk, amount);
3378 
3379 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
3380 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
3381 
3382 	if (sk_under_global_memory_pressure(sk) &&
3383 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
3384 		sk_leave_memory_pressure(sk);
3385 }
3386 
3387 /**
3388  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
3389  *	@sk: socket
3390  *	@amount: number of bytes (rounded down to a PAGE_SIZE multiple)
3391  */
3392 void __sk_mem_reclaim(struct sock *sk, int amount)
3393 {
3394 	amount >>= PAGE_SHIFT;
3395 	sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT));
3396 	__sk_mem_reduce_allocated(sk, amount);
3397 }
3398 EXPORT_SYMBOL(__sk_mem_reclaim);
3399 
3400 int sk_set_peek_off(struct sock *sk, int val)
3401 {
3402 	WRITE_ONCE(sk->sk_peek_off, val);
3403 	return 0;
3404 }
3405 EXPORT_SYMBOL_GPL(sk_set_peek_off);
3406 
3407 /*
3408  * Set of default routines for initialising struct proto_ops when
3409  * the protocol does not support a particular function. In certain
3410  * cases where it makes no sense for a protocol to have a "do nothing"
3411  * function, some default processing is provided.
3412  */
3413 
3414 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
3415 {
3416 	return -EOPNOTSUPP;
3417 }
3418 EXPORT_SYMBOL(sock_no_bind);
3419 
3420 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
3421 		    int len, int flags)
3422 {
3423 	return -EOPNOTSUPP;
3424 }
3425 EXPORT_SYMBOL(sock_no_connect);
3426 
3427 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
3428 {
3429 	return -EOPNOTSUPP;
3430 }
3431 EXPORT_SYMBOL(sock_no_socketpair);
3432 
3433 int sock_no_accept(struct socket *sock, struct socket *newsock,
3434 		   struct proto_accept_arg *arg)
3435 {
3436 	return -EOPNOTSUPP;
3437 }
3438 EXPORT_SYMBOL(sock_no_accept);
3439 
3440 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
3441 		    int peer)
3442 {
3443 	return -EOPNOTSUPP;
3444 }
3445 EXPORT_SYMBOL(sock_no_getname);
3446 
3447 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
3448 {
3449 	return -EOPNOTSUPP;
3450 }
3451 EXPORT_SYMBOL(sock_no_ioctl);
3452 
3453 int sock_no_listen(struct socket *sock, int backlog)
3454 {
3455 	return -EOPNOTSUPP;
3456 }
3457 EXPORT_SYMBOL(sock_no_listen);
3458 
3459 int sock_no_shutdown(struct socket *sock, int how)
3460 {
3461 	return -EOPNOTSUPP;
3462 }
3463 EXPORT_SYMBOL(sock_no_shutdown);
3464 
3465 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
3466 {
3467 	return -EOPNOTSUPP;
3468 }
3469 EXPORT_SYMBOL(sock_no_sendmsg);
3470 
3471 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
3472 {
3473 	return -EOPNOTSUPP;
3474 }
3475 EXPORT_SYMBOL(sock_no_sendmsg_locked);
3476 
3477 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
3478 		    int flags)
3479 {
3480 	return -EOPNOTSUPP;
3481 }
3482 EXPORT_SYMBOL(sock_no_recvmsg);
3483 
3484 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
3485 {
3486 	/* Mirror missing mmap method error code */
3487 	return -ENODEV;
3488 }
3489 EXPORT_SYMBOL(sock_no_mmap);
3490 
3491 /*
3492  * When a file is received (via SCM_RIGHTS, etc), we must bump the
3493  * various sock-based usage counts.
3494  */
3495 void __receive_sock(struct file *file)
3496 {
3497 	struct socket *sock;
3498 
3499 	sock = sock_from_file(file);
3500 	if (sock) {
3501 		sock_update_netprioidx(&sock->sk->sk_cgrp_data);
3502 		sock_update_classid(&sock->sk->sk_cgrp_data);
3503 	}
3504 }
3505 
3506 /*
3507  *	Default Socket Callbacks
3508  */
3509 
3510 static void sock_def_wakeup(struct sock *sk)
3511 {
3512 	struct socket_wq *wq;
3513 
3514 	rcu_read_lock();
3515 	wq = rcu_dereference(sk->sk_wq);
3516 	if (skwq_has_sleeper(wq))
3517 		wake_up_interruptible_all(&wq->wait);
3518 	rcu_read_unlock();
3519 }
3520 
3521 static void sock_def_error_report(struct sock *sk)
3522 {
3523 	struct socket_wq *wq;
3524 
3525 	rcu_read_lock();
3526 	wq = rcu_dereference(sk->sk_wq);
3527 	if (skwq_has_sleeper(wq))
3528 		wake_up_interruptible_poll(&wq->wait, EPOLLERR);
3529 	sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR);
3530 	rcu_read_unlock();
3531 }
3532 
3533 void sock_def_readable(struct sock *sk)
3534 {
3535 	struct socket_wq *wq;
3536 
3537 	trace_sk_data_ready(sk);
3538 
3539 	rcu_read_lock();
3540 	wq = rcu_dereference(sk->sk_wq);
3541 	if (skwq_has_sleeper(wq))
3542 		wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI |
3543 						EPOLLRDNORM | EPOLLRDBAND);
3544 	sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
3545 	rcu_read_unlock();
3546 }
3547 
3548 static void sock_def_write_space(struct sock *sk)
3549 {
3550 	struct socket_wq *wq;
3551 
3552 	rcu_read_lock();
3553 
3554 	/* Do not wake up a writer until he can make "significant"
3555 	 * progress.  --DaveM
3556 	 */
3557 	if (sock_writeable(sk)) {
3558 		wq = rcu_dereference(sk->sk_wq);
3559 		if (skwq_has_sleeper(wq))
3560 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3561 						EPOLLWRNORM | EPOLLWRBAND);
3562 
3563 		/* Should agree with poll, otherwise some programs break */
3564 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3565 	}
3566 
3567 	rcu_read_unlock();
3568 }
3569 
3570 /* An optimised version of sock_def_write_space(), should only be called
3571  * for SOCK_RCU_FREE sockets under RCU read section and after putting
3572  * ->sk_wmem_alloc.
3573  */
3574 static void sock_def_write_space_wfree(struct sock *sk)
3575 {
3576 	/* Do not wake up a writer until he can make "significant"
3577 	 * progress.  --DaveM
3578 	 */
3579 	if (sock_writeable(sk)) {
3580 		struct socket_wq *wq = rcu_dereference(sk->sk_wq);
3581 
3582 		/* rely on refcount_sub from sock_wfree() */
3583 		smp_mb__after_atomic();
3584 		if (wq && waitqueue_active(&wq->wait))
3585 			wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT |
3586 						EPOLLWRNORM | EPOLLWRBAND);
3587 
3588 		/* Should agree with poll, otherwise some programs break */
3589 		sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT);
3590 	}
3591 }
3592 
3593 static void sock_def_destruct(struct sock *sk)
3594 {
3595 }
3596 
3597 void sk_send_sigurg(struct sock *sk)
3598 {
3599 	if (sk->sk_socket && sk->sk_socket->file)
3600 		if (send_sigurg(sk->sk_socket->file))
3601 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
3602 }
3603 EXPORT_SYMBOL(sk_send_sigurg);
3604 
3605 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
3606 		    unsigned long expires)
3607 {
3608 	if (!mod_timer(timer, expires))
3609 		sock_hold(sk);
3610 }
3611 EXPORT_SYMBOL(sk_reset_timer);
3612 
3613 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
3614 {
3615 	if (timer_delete(timer))
3616 		__sock_put(sk);
3617 }
3618 EXPORT_SYMBOL(sk_stop_timer);
3619 
3620 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer)
3621 {
3622 	if (timer_delete_sync(timer))
3623 		__sock_put(sk);
3624 }
3625 EXPORT_SYMBOL(sk_stop_timer_sync);
3626 
3627 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid)
3628 {
3629 	sk_init_common(sk);
3630 	sk->sk_send_head	=	NULL;
3631 
3632 	timer_setup(&sk->sk_timer, NULL, 0);
3633 
3634 	sk->sk_allocation	=	GFP_KERNEL;
3635 	sk->sk_rcvbuf		=	READ_ONCE(sysctl_rmem_default);
3636 	sk->sk_sndbuf		=	READ_ONCE(sysctl_wmem_default);
3637 	sk->sk_state		=	TCP_CLOSE;
3638 	sk->sk_use_task_frag	=	true;
3639 	sk_set_socket(sk, sock);
3640 
3641 	sock_set_flag(sk, SOCK_ZAPPED);
3642 
3643 	if (sock) {
3644 		sk->sk_type	=	sock->type;
3645 		RCU_INIT_POINTER(sk->sk_wq, &sock->wq);
3646 		sock->sk	=	sk;
3647 	} else {
3648 		RCU_INIT_POINTER(sk->sk_wq, NULL);
3649 	}
3650 	sk->sk_uid	=	uid;
3651 
3652 	sk->sk_state_change	=	sock_def_wakeup;
3653 	sk->sk_data_ready	=	sock_def_readable;
3654 	sk->sk_write_space	=	sock_def_write_space;
3655 	sk->sk_error_report	=	sock_def_error_report;
3656 	sk->sk_destruct		=	sock_def_destruct;
3657 
3658 	sk->sk_frag.page	=	NULL;
3659 	sk->sk_frag.offset	=	0;
3660 	sk->sk_peek_off		=	-1;
3661 
3662 	sk->sk_peer_pid 	=	NULL;
3663 	sk->sk_peer_cred	=	NULL;
3664 	spin_lock_init(&sk->sk_peer_lock);
3665 
3666 	sk->sk_write_pending	=	0;
3667 	sk->sk_rcvlowat		=	1;
3668 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
3669 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
3670 
3671 	sk->sk_stamp = SK_DEFAULT_STAMP;
3672 #if BITS_PER_LONG==32
3673 	seqlock_init(&sk->sk_stamp_seq);
3674 #endif
3675 	atomic_set(&sk->sk_zckey, 0);
3676 
3677 #ifdef CONFIG_NET_RX_BUSY_POLL
3678 	sk->sk_napi_id		=	0;
3679 	sk->sk_ll_usec		=	READ_ONCE(sysctl_net_busy_read);
3680 #endif
3681 
3682 	sk->sk_max_pacing_rate = ~0UL;
3683 	sk->sk_pacing_rate = ~0UL;
3684 	WRITE_ONCE(sk->sk_pacing_shift, 10);
3685 	sk->sk_incoming_cpu = -1;
3686 
3687 	sk_rx_queue_clear(sk);
3688 	/*
3689 	 * Before updating sk_refcnt, we must commit prior changes to memory
3690 	 * (Documentation/RCU/rculist_nulls.rst for details)
3691 	 */
3692 	smp_wmb();
3693 	refcount_set(&sk->sk_refcnt, 1);
3694 	atomic_set(&sk->sk_drops, 0);
3695 }
3696 EXPORT_SYMBOL(sock_init_data_uid);
3697 
3698 void sock_init_data(struct socket *sock, struct sock *sk)
3699 {
3700 	kuid_t uid = sock ?
3701 		SOCK_INODE(sock)->i_uid :
3702 		make_kuid(sock_net(sk)->user_ns, 0);
3703 
3704 	sock_init_data_uid(sock, sk, uid);
3705 }
3706 EXPORT_SYMBOL(sock_init_data);
3707 
3708 void lock_sock_nested(struct sock *sk, int subclass)
3709 {
3710 	/* The sk_lock has mutex_lock() semantics here. */
3711 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
3712 
3713 	might_sleep();
3714 	spin_lock_bh(&sk->sk_lock.slock);
3715 	if (sock_owned_by_user_nocheck(sk))
3716 		__lock_sock(sk);
3717 	sk->sk_lock.owned = 1;
3718 	spin_unlock_bh(&sk->sk_lock.slock);
3719 }
3720 EXPORT_SYMBOL(lock_sock_nested);
3721 
3722 void release_sock(struct sock *sk)
3723 {
3724 	spin_lock_bh(&sk->sk_lock.slock);
3725 	if (sk->sk_backlog.tail)
3726 		__release_sock(sk);
3727 
3728 	if (sk->sk_prot->release_cb)
3729 		INDIRECT_CALL_INET_1(sk->sk_prot->release_cb,
3730 				     tcp_release_cb, sk);
3731 
3732 	sock_release_ownership(sk);
3733 	if (waitqueue_active(&sk->sk_lock.wq))
3734 		wake_up(&sk->sk_lock.wq);
3735 	spin_unlock_bh(&sk->sk_lock.slock);
3736 }
3737 EXPORT_SYMBOL(release_sock);
3738 
3739 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock)
3740 {
3741 	might_sleep();
3742 	spin_lock_bh(&sk->sk_lock.slock);
3743 
3744 	if (!sock_owned_by_user_nocheck(sk)) {
3745 		/*
3746 		 * Fast path return with bottom halves disabled and
3747 		 * sock::sk_lock.slock held.
3748 		 *
3749 		 * The 'mutex' is not contended and holding
3750 		 * sock::sk_lock.slock prevents all other lockers to
3751 		 * proceed so the corresponding unlock_sock_fast() can
3752 		 * avoid the slow path of release_sock() completely and
3753 		 * just release slock.
3754 		 *
3755 		 * From a semantical POV this is equivalent to 'acquiring'
3756 		 * the 'mutex', hence the corresponding lockdep
3757 		 * mutex_release() has to happen in the fast path of
3758 		 * unlock_sock_fast().
3759 		 */
3760 		return false;
3761 	}
3762 
3763 	__lock_sock(sk);
3764 	sk->sk_lock.owned = 1;
3765 	__acquire(&sk->sk_lock.slock);
3766 	spin_unlock_bh(&sk->sk_lock.slock);
3767 	return true;
3768 }
3769 EXPORT_SYMBOL(__lock_sock_fast);
3770 
3771 int sock_gettstamp(struct socket *sock, void __user *userstamp,
3772 		   bool timeval, bool time32)
3773 {
3774 	struct sock *sk = sock->sk;
3775 	struct timespec64 ts;
3776 
3777 	sock_enable_timestamp(sk, SOCK_TIMESTAMP);
3778 	ts = ktime_to_timespec64(sock_read_timestamp(sk));
3779 	if (ts.tv_sec == -1)
3780 		return -ENOENT;
3781 	if (ts.tv_sec == 0) {
3782 		ktime_t kt = ktime_get_real();
3783 		sock_write_timestamp(sk, kt);
3784 		ts = ktime_to_timespec64(kt);
3785 	}
3786 
3787 	if (timeval)
3788 		ts.tv_nsec /= 1000;
3789 
3790 #ifdef CONFIG_COMPAT_32BIT_TIME
3791 	if (time32)
3792 		return put_old_timespec32(&ts, userstamp);
3793 #endif
3794 #ifdef CONFIG_SPARC64
3795 	/* beware of padding in sparc64 timeval */
3796 	if (timeval && !in_compat_syscall()) {
3797 		struct __kernel_old_timeval __user tv = {
3798 			.tv_sec = ts.tv_sec,
3799 			.tv_usec = ts.tv_nsec,
3800 		};
3801 		if (copy_to_user(userstamp, &tv, sizeof(tv)))
3802 			return -EFAULT;
3803 		return 0;
3804 	}
3805 #endif
3806 	return put_timespec64(&ts, userstamp);
3807 }
3808 EXPORT_SYMBOL(sock_gettstamp);
3809 
3810 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag)
3811 {
3812 	if (!sock_flag(sk, flag)) {
3813 		unsigned long previous_flags = sk->sk_flags;
3814 
3815 		sock_set_flag(sk, flag);
3816 		/*
3817 		 * we just set one of the two flags which require net
3818 		 * time stamping, but time stamping might have been on
3819 		 * already because of the other one
3820 		 */
3821 		if (sock_needs_netstamp(sk) &&
3822 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
3823 			net_enable_timestamp();
3824 	}
3825 }
3826 
3827 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
3828 		       int level, int type)
3829 {
3830 	struct sock_exterr_skb *serr;
3831 	struct sk_buff *skb;
3832 	int copied, err;
3833 
3834 	err = -EAGAIN;
3835 	skb = sock_dequeue_err_skb(sk);
3836 	if (skb == NULL)
3837 		goto out;
3838 
3839 	copied = skb->len;
3840 	if (copied > len) {
3841 		msg->msg_flags |= MSG_TRUNC;
3842 		copied = len;
3843 	}
3844 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
3845 	if (err)
3846 		goto out_free_skb;
3847 
3848 	sock_recv_timestamp(msg, sk, skb);
3849 
3850 	serr = SKB_EXT_ERR(skb);
3851 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
3852 
3853 	msg->msg_flags |= MSG_ERRQUEUE;
3854 	err = copied;
3855 
3856 out_free_skb:
3857 	kfree_skb(skb);
3858 out:
3859 	return err;
3860 }
3861 EXPORT_SYMBOL(sock_recv_errqueue);
3862 
3863 /*
3864  *	Get a socket option on an socket.
3865  *
3866  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
3867  *	asynchronous errors should be reported by getsockopt. We assume
3868  *	this means if you specify SO_ERROR (otherwise what is the point of it).
3869  */
3870 int sock_common_getsockopt(struct socket *sock, int level, int optname,
3871 			   char __user *optval, int __user *optlen)
3872 {
3873 	struct sock *sk = sock->sk;
3874 
3875 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3876 	return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen);
3877 }
3878 EXPORT_SYMBOL(sock_common_getsockopt);
3879 
3880 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
3881 			int flags)
3882 {
3883 	struct sock *sk = sock->sk;
3884 	int addr_len = 0;
3885 	int err;
3886 
3887 	err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len);
3888 	if (err >= 0)
3889 		msg->msg_namelen = addr_len;
3890 	return err;
3891 }
3892 EXPORT_SYMBOL(sock_common_recvmsg);
3893 
3894 /*
3895  *	Set socket options on an inet socket.
3896  */
3897 int sock_common_setsockopt(struct socket *sock, int level, int optname,
3898 			   sockptr_t optval, unsigned int optlen)
3899 {
3900 	struct sock *sk = sock->sk;
3901 
3902 	/* IPV6_ADDRFORM can change sk->sk_prot under us. */
3903 	return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen);
3904 }
3905 EXPORT_SYMBOL(sock_common_setsockopt);
3906 
3907 void sk_common_release(struct sock *sk)
3908 {
3909 	if (sk->sk_prot->destroy)
3910 		sk->sk_prot->destroy(sk);
3911 
3912 	/*
3913 	 * Observation: when sk_common_release is called, processes have
3914 	 * no access to socket. But net still has.
3915 	 * Step one, detach it from networking:
3916 	 *
3917 	 * A. Remove from hash tables.
3918 	 */
3919 
3920 	sk->sk_prot->unhash(sk);
3921 
3922 	/*
3923 	 * In this point socket cannot receive new packets, but it is possible
3924 	 * that some packets are in flight because some CPU runs receiver and
3925 	 * did hash table lookup before we unhashed socket. They will achieve
3926 	 * receive queue and will be purged by socket destructor.
3927 	 *
3928 	 * Also we still have packets pending on receive queue and probably,
3929 	 * our own packets waiting in device queues. sock_destroy will drain
3930 	 * receive queue, but transmitted packets will delay socket destruction
3931 	 * until the last reference will be released.
3932 	 */
3933 
3934 	sock_orphan(sk);
3935 
3936 	xfrm_sk_free_policy(sk);
3937 
3938 	sock_put(sk);
3939 }
3940 EXPORT_SYMBOL(sk_common_release);
3941 
3942 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3943 {
3944 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3945 
3946 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3947 	mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf);
3948 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3949 	mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf);
3950 	mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc);
3951 	mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued);
3952 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3953 	mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len);
3954 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3955 }
3956 
3957 #ifdef CONFIG_PROC_FS
3958 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3959 
3960 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3961 {
3962 	int cpu, idx = prot->inuse_idx;
3963 	int res = 0;
3964 
3965 	for_each_possible_cpu(cpu)
3966 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3967 
3968 	return res >= 0 ? res : 0;
3969 }
3970 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3971 
3972 int sock_inuse_get(struct net *net)
3973 {
3974 	int cpu, res = 0;
3975 
3976 	for_each_possible_cpu(cpu)
3977 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->all;
3978 
3979 	return res;
3980 }
3981 
3982 EXPORT_SYMBOL_GPL(sock_inuse_get);
3983 
3984 static int __net_init sock_inuse_init_net(struct net *net)
3985 {
3986 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3987 	if (net->core.prot_inuse == NULL)
3988 		return -ENOMEM;
3989 	return 0;
3990 }
3991 
3992 static void __net_exit sock_inuse_exit_net(struct net *net)
3993 {
3994 	free_percpu(net->core.prot_inuse);
3995 }
3996 
3997 static struct pernet_operations net_inuse_ops = {
3998 	.init = sock_inuse_init_net,
3999 	.exit = sock_inuse_exit_net,
4000 };
4001 
4002 static __init int net_inuse_init(void)
4003 {
4004 	if (register_pernet_subsys(&net_inuse_ops))
4005 		panic("Cannot initialize net inuse counters");
4006 
4007 	return 0;
4008 }
4009 
4010 core_initcall(net_inuse_init);
4011 
4012 static int assign_proto_idx(struct proto *prot)
4013 {
4014 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
4015 
4016 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) {
4017 		pr_err("PROTO_INUSE_NR exhausted\n");
4018 		return -ENOSPC;
4019 	}
4020 
4021 	set_bit(prot->inuse_idx, proto_inuse_idx);
4022 	return 0;
4023 }
4024 
4025 static void release_proto_idx(struct proto *prot)
4026 {
4027 	if (prot->inuse_idx != PROTO_INUSE_NR)
4028 		clear_bit(prot->inuse_idx, proto_inuse_idx);
4029 }
4030 #else
4031 static inline int assign_proto_idx(struct proto *prot)
4032 {
4033 	return 0;
4034 }
4035 
4036 static inline void release_proto_idx(struct proto *prot)
4037 {
4038 }
4039 
4040 #endif
4041 
4042 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot)
4043 {
4044 	if (!twsk_prot)
4045 		return;
4046 	kfree(twsk_prot->twsk_slab_name);
4047 	twsk_prot->twsk_slab_name = NULL;
4048 	kmem_cache_destroy(twsk_prot->twsk_slab);
4049 	twsk_prot->twsk_slab = NULL;
4050 }
4051 
4052 static int tw_prot_init(const struct proto *prot)
4053 {
4054 	struct timewait_sock_ops *twsk_prot = prot->twsk_prot;
4055 
4056 	if (!twsk_prot)
4057 		return 0;
4058 
4059 	twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s",
4060 					      prot->name);
4061 	if (!twsk_prot->twsk_slab_name)
4062 		return -ENOMEM;
4063 
4064 	twsk_prot->twsk_slab =
4065 		kmem_cache_create(twsk_prot->twsk_slab_name,
4066 				  twsk_prot->twsk_obj_size, 0,
4067 				  SLAB_ACCOUNT | prot->slab_flags,
4068 				  NULL);
4069 	if (!twsk_prot->twsk_slab) {
4070 		pr_crit("%s: Can't create timewait sock SLAB cache!\n",
4071 			prot->name);
4072 		return -ENOMEM;
4073 	}
4074 
4075 	return 0;
4076 }
4077 
4078 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
4079 {
4080 	if (!rsk_prot)
4081 		return;
4082 	kfree(rsk_prot->slab_name);
4083 	rsk_prot->slab_name = NULL;
4084 	kmem_cache_destroy(rsk_prot->slab);
4085 	rsk_prot->slab = NULL;
4086 }
4087 
4088 static int req_prot_init(const struct proto *prot)
4089 {
4090 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
4091 
4092 	if (!rsk_prot)
4093 		return 0;
4094 
4095 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
4096 					prot->name);
4097 	if (!rsk_prot->slab_name)
4098 		return -ENOMEM;
4099 
4100 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
4101 					   rsk_prot->obj_size, 0,
4102 					   SLAB_ACCOUNT | prot->slab_flags,
4103 					   NULL);
4104 
4105 	if (!rsk_prot->slab) {
4106 		pr_crit("%s: Can't create request sock SLAB cache!\n",
4107 			prot->name);
4108 		return -ENOMEM;
4109 	}
4110 	return 0;
4111 }
4112 
4113 int proto_register(struct proto *prot, int alloc_slab)
4114 {
4115 	int ret = -ENOBUFS;
4116 
4117 	if (prot->memory_allocated && !prot->sysctl_mem) {
4118 		pr_err("%s: missing sysctl_mem\n", prot->name);
4119 		return -EINVAL;
4120 	}
4121 	if (prot->memory_allocated && !prot->per_cpu_fw_alloc) {
4122 		pr_err("%s: missing per_cpu_fw_alloc\n", prot->name);
4123 		return -EINVAL;
4124 	}
4125 	if (alloc_slab) {
4126 		prot->slab = kmem_cache_create_usercopy(prot->name,
4127 					prot->obj_size, 0,
4128 					SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT |
4129 					prot->slab_flags,
4130 					prot->useroffset, prot->usersize,
4131 					NULL);
4132 
4133 		if (prot->slab == NULL) {
4134 			pr_crit("%s: Can't create sock SLAB cache!\n",
4135 				prot->name);
4136 			goto out;
4137 		}
4138 
4139 		if (req_prot_init(prot))
4140 			goto out_free_request_sock_slab;
4141 
4142 		if (tw_prot_init(prot))
4143 			goto out_free_timewait_sock_slab;
4144 	}
4145 
4146 	mutex_lock(&proto_list_mutex);
4147 	ret = assign_proto_idx(prot);
4148 	if (ret) {
4149 		mutex_unlock(&proto_list_mutex);
4150 		goto out_free_timewait_sock_slab;
4151 	}
4152 	list_add(&prot->node, &proto_list);
4153 	mutex_unlock(&proto_list_mutex);
4154 	return ret;
4155 
4156 out_free_timewait_sock_slab:
4157 	if (alloc_slab)
4158 		tw_prot_cleanup(prot->twsk_prot);
4159 out_free_request_sock_slab:
4160 	if (alloc_slab) {
4161 		req_prot_cleanup(prot->rsk_prot);
4162 
4163 		kmem_cache_destroy(prot->slab);
4164 		prot->slab = NULL;
4165 	}
4166 out:
4167 	return ret;
4168 }
4169 EXPORT_SYMBOL(proto_register);
4170 
4171 void proto_unregister(struct proto *prot)
4172 {
4173 	mutex_lock(&proto_list_mutex);
4174 	release_proto_idx(prot);
4175 	list_del(&prot->node);
4176 	mutex_unlock(&proto_list_mutex);
4177 
4178 	kmem_cache_destroy(prot->slab);
4179 	prot->slab = NULL;
4180 
4181 	req_prot_cleanup(prot->rsk_prot);
4182 	tw_prot_cleanup(prot->twsk_prot);
4183 }
4184 EXPORT_SYMBOL(proto_unregister);
4185 
4186 int sock_load_diag_module(int family, int protocol)
4187 {
4188 	if (!protocol) {
4189 		if (!sock_is_registered(family))
4190 			return -ENOENT;
4191 
4192 		return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK,
4193 				      NETLINK_SOCK_DIAG, family);
4194 	}
4195 
4196 #ifdef CONFIG_INET
4197 	if (family == AF_INET &&
4198 	    protocol != IPPROTO_RAW &&
4199 	    protocol < MAX_INET_PROTOS &&
4200 	    !rcu_access_pointer(inet_protos[protocol]))
4201 		return -ENOENT;
4202 #endif
4203 
4204 	return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK,
4205 			      NETLINK_SOCK_DIAG, family, protocol);
4206 }
4207 EXPORT_SYMBOL(sock_load_diag_module);
4208 
4209 #ifdef CONFIG_PROC_FS
4210 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
4211 	__acquires(proto_list_mutex)
4212 {
4213 	mutex_lock(&proto_list_mutex);
4214 	return seq_list_start_head(&proto_list, *pos);
4215 }
4216 
4217 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4218 {
4219 	return seq_list_next(v, &proto_list, pos);
4220 }
4221 
4222 static void proto_seq_stop(struct seq_file *seq, void *v)
4223 	__releases(proto_list_mutex)
4224 {
4225 	mutex_unlock(&proto_list_mutex);
4226 }
4227 
4228 static char proto_method_implemented(const void *method)
4229 {
4230 	return method == NULL ? 'n' : 'y';
4231 }
4232 static long sock_prot_memory_allocated(struct proto *proto)
4233 {
4234 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
4235 }
4236 
4237 static const char *sock_prot_memory_pressure(struct proto *proto)
4238 {
4239 	return proto->memory_pressure != NULL ?
4240 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
4241 }
4242 
4243 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
4244 {
4245 
4246 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
4247 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
4248 		   proto->name,
4249 		   proto->obj_size,
4250 		   sock_prot_inuse_get(seq_file_net(seq), proto),
4251 		   sock_prot_memory_allocated(proto),
4252 		   sock_prot_memory_pressure(proto),
4253 		   proto->max_header,
4254 		   proto->slab == NULL ? "no" : "yes",
4255 		   module_name(proto->owner),
4256 		   proto_method_implemented(proto->close),
4257 		   proto_method_implemented(proto->connect),
4258 		   proto_method_implemented(proto->disconnect),
4259 		   proto_method_implemented(proto->accept),
4260 		   proto_method_implemented(proto->ioctl),
4261 		   proto_method_implemented(proto->init),
4262 		   proto_method_implemented(proto->destroy),
4263 		   proto_method_implemented(proto->shutdown),
4264 		   proto_method_implemented(proto->setsockopt),
4265 		   proto_method_implemented(proto->getsockopt),
4266 		   proto_method_implemented(proto->sendmsg),
4267 		   proto_method_implemented(proto->recvmsg),
4268 		   proto_method_implemented(proto->bind),
4269 		   proto_method_implemented(proto->backlog_rcv),
4270 		   proto_method_implemented(proto->hash),
4271 		   proto_method_implemented(proto->unhash),
4272 		   proto_method_implemented(proto->get_port),
4273 		   proto_method_implemented(proto->enter_memory_pressure));
4274 }
4275 
4276 static int proto_seq_show(struct seq_file *seq, void *v)
4277 {
4278 	if (v == &proto_list)
4279 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
4280 			   "protocol",
4281 			   "size",
4282 			   "sockets",
4283 			   "memory",
4284 			   "press",
4285 			   "maxhdr",
4286 			   "slab",
4287 			   "module",
4288 			   "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n");
4289 	else
4290 		proto_seq_printf(seq, list_entry(v, struct proto, node));
4291 	return 0;
4292 }
4293 
4294 static const struct seq_operations proto_seq_ops = {
4295 	.start  = proto_seq_start,
4296 	.next   = proto_seq_next,
4297 	.stop   = proto_seq_stop,
4298 	.show   = proto_seq_show,
4299 };
4300 
4301 static __net_init int proto_init_net(struct net *net)
4302 {
4303 	if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops,
4304 			sizeof(struct seq_net_private)))
4305 		return -ENOMEM;
4306 
4307 	return 0;
4308 }
4309 
4310 static __net_exit void proto_exit_net(struct net *net)
4311 {
4312 	remove_proc_entry("protocols", net->proc_net);
4313 }
4314 
4315 
4316 static __net_initdata struct pernet_operations proto_net_ops = {
4317 	.init = proto_init_net,
4318 	.exit = proto_exit_net,
4319 };
4320 
4321 static int __init proto_init(void)
4322 {
4323 	return register_pernet_subsys(&proto_net_ops);
4324 }
4325 
4326 subsys_initcall(proto_init);
4327 
4328 #endif /* PROC_FS */
4329 
4330 #ifdef CONFIG_NET_RX_BUSY_POLL
4331 bool sk_busy_loop_end(void *p, unsigned long start_time)
4332 {
4333 	struct sock *sk = p;
4334 
4335 	if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
4336 		return true;
4337 
4338 	if (sk_is_udp(sk) &&
4339 	    !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
4340 		return true;
4341 
4342 	return sk_busy_loop_timeout(sk, start_time);
4343 }
4344 EXPORT_SYMBOL(sk_busy_loop_end);
4345 #endif /* CONFIG_NET_RX_BUSY_POLL */
4346 
4347 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len)
4348 {
4349 	if (!sk->sk_prot->bind_add)
4350 		return -EOPNOTSUPP;
4351 	return sk->sk_prot->bind_add(sk, addr, addr_len);
4352 }
4353 EXPORT_SYMBOL(sock_bind_add);
4354 
4355 /* Copy 'size' bytes from userspace and return `size` back to userspace */
4356 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
4357 		     void __user *arg, void *karg, size_t size)
4358 {
4359 	int ret;
4360 
4361 	if (copy_from_user(karg, arg, size))
4362 		return -EFAULT;
4363 
4364 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
4365 	if (ret)
4366 		return ret;
4367 
4368 	if (copy_to_user(arg, karg, size))
4369 		return -EFAULT;
4370 
4371 	return 0;
4372 }
4373 EXPORT_SYMBOL(sock_ioctl_inout);
4374 
4375 /* This is the most common ioctl prep function, where the result (4 bytes) is
4376  * copied back to userspace if the ioctl() returns successfully. No input is
4377  * copied from userspace as input argument.
4378  */
4379 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg)
4380 {
4381 	int ret, karg = 0;
4382 
4383 	ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg);
4384 	if (ret)
4385 		return ret;
4386 
4387 	return put_user(karg, (int __user *)arg);
4388 }
4389 
4390 /* A wrapper around sock ioctls, which copies the data from userspace
4391  * (depending on the protocol/ioctl), and copies back the result to userspace.
4392  * The main motivation for this function is to pass kernel memory to the
4393  * protocol ioctl callbacks, instead of userspace memory.
4394  */
4395 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
4396 {
4397 	int rc = 1;
4398 
4399 	if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET)
4400 		rc = ipmr_sk_ioctl(sk, cmd, arg);
4401 	else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6)
4402 		rc = ip6mr_sk_ioctl(sk, cmd, arg);
4403 	else if (sk_is_phonet(sk))
4404 		rc = phonet_sk_ioctl(sk, cmd, arg);
4405 
4406 	/* If ioctl was processed, returns its value */
4407 	if (rc <= 0)
4408 		return rc;
4409 
4410 	/* Otherwise call the default handler */
4411 	return sock_ioctl_out(sk, cmd, arg);
4412 }
4413 EXPORT_SYMBOL(sk_ioctl);
4414 
4415 static int __init sock_struct_check(void)
4416 {
4417 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops);
4418 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off);
4419 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue);
4420 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue);
4421 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog);
4422 
4423 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst);
4424 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex);
4425 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie);
4426 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf);
4427 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter);
4428 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq);
4429 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready);
4430 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo);
4431 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat);
4432 
4433 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err);
4434 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket);
4435 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg);
4436 
4437 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock);
4438 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem);
4439 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc);
4440 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags);
4441 
4442 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4443 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc);
4444 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf);
4445 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued);
4446 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc);
4447 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags);
4448 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head);
4449 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue);
4450 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending);
4451 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm);
4452 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status);
4453 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag);
4454 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer);
4455 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate);
4456 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey);
4457 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey);
4458 
4459 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate);
4460 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo);
4461 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority);
4462 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark);
4463 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache);
4464 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps);
4465 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type);
4466 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size);
4467 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation);
4468 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash);
4469 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs);
4470 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift);
4471 	CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag);
4472 	return 0;
4473 }
4474 
4475 core_initcall(sock_struct_check);
4476